AI and Cancer Research ── Lymphoma ── 2026-11-05
Lymphoma: treatment, trials and AI
- Abstract
To develop and validate an interpretable multimodal model integrating deep learning radiomics with clinical data using baseline [ 18 F] FDG PET/CT for non-invasive identification of the double-expressor lymphoma (DEL) in diffuse large B-cell lymphoma (DLBCL). The histopathological reference standard for DLBCL diagnosis and DEL subtyping was immunohistochemistry according to the 2016 WHO. Patients from January 2019 to July 2025 were retrospectively enrolled and randomly divided into training and test sets at 7:3, and a patient-level analysis was performed. Clinical, radiomics and deep learning features were extracted. After LASSO selection, individual and fusion models were constructed. Statistical analyses were performed using Python (3.6), and p-value < 0.05 was considered significant. Model performance was assessed via receiver operating characteristic (ROC), calibration and decision curve analysis (DCA). SHAP and a nomogram were used for interpretation. A total of 143 patients (82 males, 61 females, median age 61.5 years) were enrolled. ViT_KNN model achieved a test AUC of 0.936 (95%Cl 0.856-1.000), outperforming clinical (AUC 0.857, 95%Cl 0.722-0.993) and radiomics (AUC 0.706, 95%Cl 0.547-0.865) models. Late fusion, especially Stacking_LR, achieved the best performance (AUC 0.948, 95%Cl 0.865-1.000, accuracy 95.3%, specificity 100%, sensitivity 88.2%). Calibration and DCA confirmed clinical net benefit. SHAP analysis highlighted deep learning features as most influential. The Stacking_LR nomogram offers an intuitive clinical tool. Deep learning radiomics based on baseline [ 18 F] FDG PET/CT enables effective non-invasive identification of DEL. The stacking ensemble strategy enhances multimodal fusion and provides a highly accurate, interpretable tool for DLBCL risk stratification.
Journal IF-equivalent: 3.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Liang J, Lu G, Zhong Q, Peng Q, Tian R, Zheng W, et al. Prediction of Double-Expressor Lymphoma in Diffuse Large B-Cell Lymphoma Using a Non-invasive Model Based on [18F] FDG PET/CT and Stacking. Acad Radiol. 2026 Oct [Epub ahead of print]. doi:10.1016/j.acra.2026.08.091. PMID: 42934072.Checked: Abstract only - Abstract
Accurate subtyping of lymphomas is critical for guiding treatment decisions and predicting patient outcomes, yet it remains challenging due to overlapping morphological features. This study aimed to develop and validate a deep learning-based diagnostic system, designated BH-LymphDS, to accurately differentiate the three most prevalent lymphoma subtypes: Diffuse Large B-cell Lymphoma (DLBCL), Follicular Lymphoma (FL), and Hodgkin Lymphoma (HL). The system integrates a Swin Transformer backbone with a clustering-constrained attention multiple instance learning (CLAM) algorithm and a k-nearest neighbors (KNN) classifier, supporting weakly supervised learning directly from whole-slide images without manual region-level annotation. A total of 1910 whole-slide images (WSIs) from diagnostic biopsies were utilized for model development-the largest cohort to date in this field. The performance of nine distinct deep learning architectures was systematically compared. Model efficacy was rigorously evaluated through five-fold cross-validation and further validated on independent external datasets from TCIA and TCGA. Additionally, a comparative analysis was conducted against the diagnostic performance of three experienced hematopathologists. In the internal testing cohort, BH-LymphDS achieved a micro-average AUC of 0.955 and an accuracy of 92.5%. During cross-validation, the system attained a superior micro-average F1-score (0.871) compared to the average of the three hematopathologists (0.851). The Swin Transformer emerged as the optimal backbone, and gradient-weighted class activation mapping (GradCAM) confirmed that the model's decisions were based on legitimate morphological features such as nuclear atypia and cellular growth patterns. Our findings demonstrate that an advanced AI system can satisfy the rigorous requirements for accurate lymphoma subclassification on histopathological slides. The BH-LymphDS shows strong potential as a reliable auxiliary diagnostic tool to enhance clinical workflow efficiency and diagnostic consistency.
Journal IF-equivalent: 5.3 (OpenAlex 2-year mean citedness, value as of 2026-10-07, retrieved 2026-10-08; not the official Clarivate IF)Reference: Zhang P, Yan C, Sun M, Ding S, Li L, Cao W, et al. Deep Learning on Histopathological Images Enables Accurate Diagnosis of Common B-Cell and Hodgkin Lymphoma. Cancers. 2026 Oct 8;18(19):3237. doi:10.3390/cancers18193237.Checked: Abstract only - Abstract
Deep learning (DL) methods have shown strong performance in predicting dose distributions from patient anatomy and target structures; however, accurate dose prediction does not guarantee a deliverable plan after conversion into machine parameters and dose recalculation in the treatment planning system (TPS), motivating training objectives that explicitly couple prediction to deliverability. Lymphoma presents substantial clinical heterogeneity in target location, extent, and prescription/fractionation, so fast automated methods that produce deliverable plans are particularly valuable. To investigate whether a deliverability-aware training framework improves the end-to-end deliverability of DL-generated intensity-modulated radiotherapy (IMRT) plans for lymphoma. A retrospective single-institution dataset of 619 plans (571 patients; 2009-2021) was used. Clinical volumetric modulated arc therapy (VMAT) plans were converted to standardized 15-field coplanar sliding-window IMRT reference plans in Eclipse using automated scripting. Two DL approaches were compared: a baseline dose-only objective model (DL-DO) and a multi-objective model (DL-MO) trained with dose supervision, fluence supervision, and a dose-fluence-dose cycle-consistency constraint implemented via fixed inverse and forward networks. Predicted fluence maps were imported into the TPS, leaf-sequenced, and recalculated with AcurosXB without further optimization; plan quality was evaluated on a held-out test set (n = 61) using planning target volume (PTV) D 98 , D 2 , and D max with paired Wilcoxon testing and Holm correction. DL-MO produced significantly better machine-deliverable plans than DL-DO, improving coverage and reducing hotspots (median paired differences: +12.31 percentage points (pp) in D 98 , -3.43 pp in D 2 , and -3.18 pp in D max ; Holm-corrected p < 0.0001 for all). Mean ± SD PTV metrics were D 98 = 90.10 ± 3.74% vs. 78.59 ± 5.92%, D 2 = 106.28 ± 1.13% vs. 109.65 ± 2.40%, and D max = 110.32 ± 2.20% vs. 113.27 ± 3.03% (DL-MO vs. DL-DO). DL-MO approached the TPS reference IMRT plans (D 98 = 91.62 ± 3.45%) with residual hotspot elevation. Cycle-consistency training improves the end-to-end deliverability of DL-generated IMRT plans for heterogeneous lymphoma, yielding substantially better TPS-recalculated, machine-deliverable plans than conventional dose-only supervision while maintaining sub-second fluence-map generation. However, statistically significant differences from the TPS reference remained for D 98 , D 2 , and D max .
Journal IF-equivalent: 2.4 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Kutnár D, Vogelius IR, Georgiev HA, Specht L, Petersen J. Improving deep‐learning IMRT plan quality for lymphoma using deliverability‐aware training. J Applied Clin Med Phys. 2026 Oct 5;27(10):e70815. doi:10.1002/acm2.70815.Checked: Abstract only - Exploratory Radiomic Analysis of Anterior Segment OCT Images in Patients with Vitreoretinal LymphomaAbstract
To explore the ability of a radiomics-based classification model of identifying anterior segment optical coherence tomography (AS-OCT) image features compatible with active vitreoretinal lymphoma (VRL). The secondary aim was to investigate the possible association between these image features and aqueous humour molecular biomarkers used in VRL assessment. We conducted a retrospective single-centre study of patients with biopsy-proven VRL, and at least one series of baseline or follow-up AS-OCT image scans and an aqueous tap. We extracted radiomic features from a manually delineated region of interest (ROI) in anterior vitreous AS-OCT images, five were selected and trained XGBoost classifiers with leave-two-patients-out cross-validation (L2PO) to distinguish active from inactive disease, using AUC-ROC as performance metric. If present, interleukin (IL)-10:IL-6 ratios were correlated with imaging features using logistic regression analyses. Eight patients (22 visits total, 43 eyes) were analysed. From the L2PO analysis, we obtained an AUC metric of 0.74 with a 95% confidence interval (CI) of 0.68-0.81. GLRLM_LRLGE was the feature most consistently associated with disease activity. When IL-10:IL-6 > 1 was used as the outcome, the association weakened (all p -values > 0.05), though mean and median values showed the same direction of effect. This pilot study explored the potential innovative role of radiomics to analyse VRL activity status using AS-OCT images. Although it is currently far from being a clinical tool, radiomic analysis could be the subject of future investigations aiming to refine non-invasive monitoring strategies. Prospective studies with standardized, simultaneous serial imaging and aqueous tap are required.
Journal IF-equivalent: 1.8 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Gozzi F, Bertolini M, Vicentini M, Ragusa E, Trojani V, Mancuso P, et al. Exploratory Radiomic Analysis of Anterior Segment OCT Images in Patients with Vitreoretinal Lymphoma. Ocular Immunology and Inflammation. 2026 Oct 5 [Epub ahead of print]. doi:10.1080/09273948.2026.2739454.Checked: Abstract only - Abstract
Background/Objectives : To evaluate a deep learning-based method for automatic segmentation of Hodgkin lymphoma (HL) on PET images and to explore the prognostic value of total metabolic tumor volume (TMTV) derived from segmentation. Methods : PET datasets from two independent institutions were retrospectively collected and divided into training and validation cohorts. MedNeXt was trained for HL PET segmentation and benchmarked against five representative segmentation architectures under the same data split, preprocessing pipeline, and evaluation framework. Segmentation performance was assessed using the Dice similarity coefficient (DSC), Jaccard index (JSC), 95th percentile Hausdorff distance (HD95), and average symmetric surface distance (ASSD). Bland-Altman and linear regression analyses evaluated agreement between ground-truth TMTV (gtTMTV) and predicted TMTV (pTMTV). Cox proportional hazards regression analysis was applied to assess the prognostic value of pTMTV. Results : In the validation cohort, the mean values (±standard deviation) of DSC, HD95, JSC, and ASSD were 0.702 ± 0.191, 67.998 ± 99.220, 0.561 ± 0.203, and 16.992 ± 31.659, respectively. Among the evaluated segmentation models, MedNeXt showed the best segmentation performance. Mean gtTMTV and pTMTV were 424.681 ± 498.811 cm 3 and 405.871 ± 466.984 cm 3 , showing good agreement. pTMTV was significantly associated with progression-free survival (PFS) ( p = 0.048), but not overall survival (OS) ( p = 0.163). Conclusions : MedNeXt achieved reliable HL segmentation, although tumor boundary delineation requires further improvement. The derived pTMTV may serve as a prognostic indicator for PFS but showed no significant association with OS. Further validation in larger cohorts is warranted.
Journal IF-equivalent: 5.3 (OpenAlex 2-year mean citedness, value as of 2026-10-07, retrieved 2026-10-08; not the official Clarivate IF)Reference: Zhang H, Sun F, Yu H, Chen X, Shi Y, Teng Y, et al. MedNeXt-Based Automated Tumor Segmentation and TMTV Prognostication in Hodgkin Lymphoma PET Images: A Retrospective Study. Cancers. 2026 Oct 3;18(19):3196. doi:10.3390/cancers18193196.Checked: Abstract only - Abstract
Hodgkin lymphoma (HL) survivors face an elevated long-term risk of developing second primary malignancies (SPMs), yet accurate prediction tools remain scarce. This study aimed to develop and internally validate a machine learning-based prediction model for SPM risk in HL survivors using a competing risk framework. We analysed 57,429 patients diagnosed with HL between 2006 and 2020 from the Surveillance, Epidemiology, and End Results (SEER) database. Patients were randomly divided into training ( n = 40,201, 70%) and internal validation ( n = 17,228, 30%) cohorts. A random survival forest (RSF) model was developed to predict SPM occurrence, with death without SPM treated as a competing event. Model performance was compared against the traditional Fine-Grey subdistribution hazard regression. Discrimination was assessed using time-dependent area under the receiver operating characteristic curve (AUC), and calibration was evaluated through calibration plots. Clinical utility was determined via decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) analysis was employed to interpret the model. During follow-up, 4,954 patients (8.6%) developed SPMs, while 9,750 (17.0%) died without SPM. The RSF model demonstrated superior discriminative ability compared to Fine-Grey regression across all time points: AUC at 3 years (0.721 vs. 0.549), 5 years (0.761 vs. 0.555), and 10 years (0.804 vs. 0.553). Calibration plots indicated good agreement between predicted and observed risks. DCA confirmed the RSF model provided greater net clinical benefit across a wide range of threshold probabilities. SHAP analysis identified age as the most influential predictor, followed by disease stage, radiotherapy status, race, sex, and chemotherapy. The RSF-based machine learning model accurately predicts SPM risk in HL survivors within a competing risk framework, substantially outperforming traditional regression approaches. This interpretable prediction tool may facilitate personalised surveillance strategies and risk-adapted follow-up for HL survivors.
Journal IF-equivalent: 2.3 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Li S, Xiong M, Gan Z, Lu H, Zhao Y, Huang H, et al. Development and internal validation of a machine learning model for predicting second primary malignancy risk in hodgkin lymphoma survivors: a population-based competing risk analysis. Hematology. 2026 Sep 28;31(1):2737562. doi:10.1080/16078454.2026.2737562. PMID: 42804656.Checked: Abstract only - Abstract
This study evaluated an AI-based lesion segmentation tool using a hybrid 2D-3D deep learning model based on the nnUNet architecture with a ResNet18 backbone, incorporating post-processing to reduce false positives and segmentation-related quantitative errors. Whole-body FDG PET/CT scans from 72 lymphoma patients (mean age 39.2 ± 18.6 years; 33 females, 39 males; 43.1% Hodgkin lymphoma, 56.9% non-Hodgkin lymphoma) acquired on a uMI550 digital PET/CT system were retrospectively analysed. Automated AI-based segmentations were compared with manual delineations by experienced physicians. Performance was assessed using F1-score, sensitivity, and positive predictive value. Quantitative agreement for all PET metrics was evaluated using Pearson correlation, concordance correlation coefficient, intraclass correlation coefficient, regression, and Bland-Altman analysis. McNemar's test assessed detection differences. The correlations between DSC and image quality metrics were examined. A total of 711 AI and 848 manually segmented lesions were analysed. The AI model achieved an excellent DSC (89.2%), sensitivity (82.9) and PPV (96.5) which indicated balanced detection performance. Strong correlations were observed for all quantitative parameters (r > 0.95, p < 0.001; CCC > 0.94; ICC > 0.94; 95% CI: 0.93-0.99), with minimal bias on Bland-Altman analysis. DSC showed a moderate negative correlation with liver SNR (r = -0.40, p < 0.001) and weak correlations with liver SUV metrics. The fully automated AI-based segmentation tool demonstrated good-to-excellent lesion detection accuracy and high quantitative concordance with manual segmentation in whole-body FDG PET/CT for lymphoma patients. Continued model refinement and region-specific training are warranted to enhance detection performance.
Journal IF-equivalent: 1.3 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Masogo T, Wadhwa P, Kaoma C, Seebarruth Y, Qebetu M, Banda K, et al. Evaluation of the detection accuracy of a fully automated AI-based lesion segmentation tool in whole-body FDG PET/CT for lymphoma. Z Med Phys. 2026 Sep [Epub ahead of print]. doi:10.1016/j.zemedi.2026.09.002. PMID: 42802086.Checked: Abstract only - Abstract
Accurate classification of lymphoma cell subtypes is essential for disease diagnosis and therapeutic decision-making, yet conventional approaches often rely on fluorescence labeling, labor-intensive sample preparation, and specialized instrumentation, limiting their applicability for rapid, label-free single-cell analysis. Here, we present an AI-assisted microfluidic light-scattering imaging platform for label-free classification of lymphoma cells. The platform integrates hydrodynamic focusing within a microfluidic chip, continuous acquisition of two-dimensional (2D) light-scattering patterns, automated image preprocessing, and transfer learning based on a pretrained ResNet50 network for intelligent optical feature extraction and classification. Human B lymphoma (Daudi) and T lymphoblastic lymphoma (SUP-T1) cells were used to evaluate the proposed framework. The optical imaging system was first validated using standard microspheres, demonstrating reliable acquisition of light-scattering patterns under continuous-flow conditions. A dataset comprising 800 single-cell scattering patterns was subsequently established and evaluated using stratified five-fold cross-validation. The proposed framework achieved an average classification accuracy of 94.75% with an average area under the receiver operating characteristic (ROC) curve of 0.986. By integrating microfluidic optical biosensing with deep learning, this work enables automated interpretation of intrinsic optical scattering signatures and provides a promising AI-enabled strategy for rapid, label-free lymphoma screening and intelligent healthcare applications.
Journal IF-equivalent: 8.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Xie L, Wang M, Luo X, Xia S, Ren Q, Zhou X. Microfluidic Light-Scattering Imaging Coupled with Deep Learning for Label-Free Single-Cell Classification of Lymphoma Cells. Biosensors (Basel). 2026 Sep 7;16(9):500. doi:10.3390/bios16090500. PMID: 42783167; PMCID: PMC13604312.Checked: Full text checked - Abstract
Prediction of the outcome of large B-cell lymphomas/high-grade B-cell lymphomas (DLBCLs/HGBCLs) is based on clinical parameters and molecular testing, for example, for rearrangements of MYC (MYC-R) and MYC-R in combination with BCL2 and BCL6 translocations (double/triple hit). However, the group of DLBCL/HGBCL with poor outcome is not confined to MYC-R lymphomas, and fluorescence in situ hybridization (FISH) testing for MYC-R misses several high-risk lymphomas. We aimed to understand if artificial intelligence (AI) trained to identify MYC-R will delineate a poor prognostic subgroup. We generated a collection of digital hematoxylin and eosin (H&E)-stained slides of DLBCL/HGBCL ( N = 2018) annotated for MYC, BCL2, and BCL6 translocations. A multiple-instance deep learning AI model for the identification of MYC-R alone or as double/triple hit was established using 1035 H&E-stained slides and evaluated on an external test cohort ( N = 499). A pretrained tumor tissue classifier improved reliability and interpretability by focusing the model on tumor areas. Our model score reflects a morphological "MYCness" in DLBCL/HGBCL and demonstrates a strong association with overall survival (OS) and progression-free survival (PFS) in the external test cohort. This was confirmed with an additional clinical test cohort ( N = 484) without FISH labels. The AI model scores correlate with various molecular features of DLBCL/HGBCL, including BCL2 and MYC gene expression, and the high-grade gene expression signature, but also features of the tumor microenvironment. Multivariate analysis, adjusted for International Prognostic Index (IPI) factors, demonstrated the prognostic significance of our model in identifying high-risk cases for both PFS and OS.
Journal IF-equivalent: 3.7 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Lippl J, Reinke S, Schrod S, Wolfseher L, Hüttl P, Huttner M, et al. AI for prognostic assessment of diffuse large B‐cell lymphoma using H&E whole‐slide images. Hemasphere. 2026 Aug 31;10(9):e70463. doi:10.1002/hem3.70463. PMID: 42676952; PMCID: PMC13528628.Checked: Full text checked - Abstract
Background/Objectives : Diffuse large B-cell lymphoma (DLBCL) remains clinically heterogeneous despite standard first-line immunochemotherapy. We evaluated whether machine learning (ML) models using routinely available diagnostic variables could predict primary refractory DLBCL. Materials and Methods : We performed a retrospective single-center study of 369 patients with DLBCL treated between 2015 and 2023. Primary refractory disease was defined as failure to achieve at least a partial response after first-line therapy. Logistic regression, random forest, gradient boosting, support vector machine with radial basis function kernel (SVM-RBF), and multilayer perceptron models were trained using baseline clinical and laboratory predictors. Performance was assessed using a stratified 80/20 holdout split, nested cross-validation, repeated random splits, and sensitivity analysis with complete hyperparameter retuning. Results : Primary refractory disease occurred in 111 patients (30.1%). In the holdout test set ( n = 73), IPI achieved the highest discrimination (AUC 0.692; 95% CI 0.545-0.828). Among ML models, gradient boosting performed best (AUC 0.658; 95% CI 0.503-0.801), followed by SVM-RBF (AUC 0.619; 95% CI 0.471-0.760). DeLong comparisons showed no significant differences between ML models and IPI. In internal validation, gradient boosting showed the highest nested CV AUC (0.712 ± 0.049). Feature importance identified IPI as the dominant predictor. A weighted ML-derived clinical score identified a high-risk group with the highest refractory rate (52.9%), although the trend was not statistically significant ( p = 0.063). Conclusions : ML models showed comparable but not superior performance to IPI for predicting primary refractory DLBCL. These findings support further external validation and development of multimodal clinically interpretable prediction frameworks for risk-adapted management.
Journal IF-equivalent: 4.0 (OpenAlex 2-year mean citedness, value as of 2026-10-29, retrieved 2026-10-30; not the official Clarivate IF)Reference: Minciuna CD, Minciuna D, Vasilache IA, Miron L. Predicting Primary Refractory Diffuse Large B-Cell Lymphoma: A Comparison of Machine Learning Models with the IPI Score. J Clin Med. 2026 Aug 27;15(17):6628. doi:10.3390/jcm15176628. PMID: 42739634; PMCID: PMC13566510.Checked: Full text checked - Abstract
The clinical outcomes of diffuse large B-cell lymphoma (DLBCL) are highly heterogeneous. While clinical indices like the international prognostic index (IPI) are widely used, their predictive accuracy remains limited. The integration of molecular features with clinical characteristics holds promise for developing more precise prognostic models to improve risk stratification and personalize treatment strategies. This study aimed to systematically identify key factors influencing overall survival (OS) and relapse in patients with DLBCL by leveraging publicly available transcriptomic data and clinical information. The goal was to construct and validate a high-precision risk-prediction model by using machine learning methods to aid in individualized clinical decision-making. We curated clinical and transcriptomic data from the GSE31312 cohort. A baseline clinical model was first constructed using multivariate Cox regression. Key genes associated with prognosis were identified through univariate Cox and survival analyses. Subsequently, 3 machine learning survival models, namely, fast survival support vector machine (FastSurvivalSVM), gradient boosting survival analysis (GBSurvival), and random survival forest (RSF), were trained and evaluated using 5-fold cross-validation. The interpretability of the optimal model was further elucidated using Shapley Additive Explanations (SHAP) methodology. The baseline clinical model confirmed age, elevated lactate dehydrogenase, Eastern Cooperative Oncology Group score, Ann Arbor stage, and B symptoms as independent risk factors for OS and relapse-free survival, with a C-index of 0.65-0.67. At the molecular level, genes such as PSMG4 and CRY1 were significantly associated with poor OS, while TMEM182 and SPIRE1 were prominent in relapse prediction. Among the machine learning models, FastSurvivalSVM demonstrated the best overall performance, achieving an area under the curve of 0.791 for 1-year OS prediction and 0.774 for 1-year relapse prediction. SHAP analysis revealed that both clinical (eg, IPI and age) and molecular (eg, PSMG4 and SPIRE1) features were critical drivers of the model's predictions. This study successfully developed a multidimensional risk prediction model that integrates clinical and molecular characteristics for DLBCL. The FastSurvivalSVM model showed superior performance in predicting mortality and relapse risks. The interpretability analysis uncovered key prognostic factors, providing a valuable tool for personalized risk management and new theoretical insights for future mechanistic research.
Journal IF-equivalent: 6.7 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Zhao J, Wen X, Ma L, Su L. Integrated Clinical-Molecular Risk Stratification in Diffuse Large B-Cell Lymphoma: Machine Learning Survival Analysis. JMIR Med Inform. 2026 Aug 21;14:e84636. doi:10.2196/84636. PMID: 42627678; PMCID: PMC13545529.Checked: Full text checked - Abstract
Relapsed/refractory classical Hodgkin lymphoma (R/R cHL) remains clinically challenging due to substantial heterogeneity in relapse risk. Reliable, non-invasive tools for improved relapse risk stratification are urgently needed to guide individualized therapeutic strategies. In this multicenter retrospective study, 161 patients with R/R cHL from five institutions were included (training cohort: n = 102; validation cohort: n = 59). Clinical and metabolic covariates were assessed at the time of relapsed/refractory disease and baseline 18F-FDG PET/CT before salvage treatment. We developed a deep learning-based Mixture-of-Experts (MoE) framework that integrates four medical foundation models (PET-Diffusion, SAM-Med2D, MedCLIP, and RadFM) to derive a quantitative imaging biomarker (MoEScore) from baseline 18 F-FDG PET/CT. A multiparametric model incorporating MoEScore and independent clinical/metabolic predictors was constructed and evaluated using discrimination, calibration, and clinical utility analyses. Model interpretability was assessed using attention visualization, ablation analysis, and pathological correlation. MoEScore demonstrated predictive performance (AUC: 0.861 in training; 0.783 in validation) and remained independently associated with relapse (HR = 11.18, 95% CI: 2.48-50.45; P = 0.002). The multiparametric model achieved a C-index of 0.785 in the training cohort and 0.717 in the validation cohort, compared with 0.598-0.752 for the clinical and clinical-metabolic models. MoEScore consistently stratified relapse risk across both relapsed and refractory subgroups. Exploratory interpretability analyses indicated that model saliency was predominantly localized to metabolically active lesion regions and suggested a greater relative contribution of PET than CT. MoEScore distributions were broadly consistent with known histopathological subtype patterns, supporting further biological evaluation. This study presents an exploratory, non-invasive deep learning framework for relapse risk stratification in R/R cHL. By integrating multimodal imaging and expert-level representations, the MoE model may capture tumor heterogeneity beyond conventional metrics and may help inform risk-adapted therapeutic strategies after further validation.
Journal IF-equivalent: 5.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Jiang C, Jiang Z, Zhang X, Zhang Z, Teng Y, Zhou H, et al. Deep learning-based PET/CT mixture-of-experts model for relapse risk stratification in relapsed/refractory classical hodgkin lymphoma: a multicenter study. Eur J Nucl Med Mol Imaging. 2026 Aug 19;53(13):8358-71. doi:10.1007/s00259-026-08146-0. PMID: 42613444.Checked: Abstract only - Abstract
Accurate segmentation of diffuse large B-cell lymphoma (DLBCL) is critical for reliable PET/CT-based quantification and total metabolic tumor volume (TMTV) estimation. However, most deep learning approaches rely on limited annotated datasets and conventional convolutional architectures, which may restrict generalization and training stability. In this study, we propose a multimodal PET/CT Swin UNETR framework enhanced through large-scale self-supervised pretraining. A total of 717 FDG-PET/CT scans were collected from a single center, including 500 unlabeled volumes used for self-supervised pretraining and 217 expert-annotated cases for supervised learning. The labeled dataset was randomly divided into 174 training cases and an independent held-out test cohort of 43 cases. A standardized preprocessing pipeline with isotropic resampling and voxel harmonization was applied to ensure spatial consistency. The proposed SSL + PET/CT Swin UNETR achieved a Dice score of 0.723 (95% CI: 0.70-0.75), IoU of 0.648, HD of 15.0 mm, and HD95 of 8.8 mm on a held-out internal test set, evaluated supervised baseline models (p < 0.001, Wilcoxon signed-rank test). TMTV regression analysis demonstrated strong correlation with ground-truth measurements (R² = 0.929) with minimal systematic bias in Bland-Altman evaluation. These findings demonstrate the potential of multimodal self-supervised learning to improve automated DLBCL lesion segmentation in PET/CT images and support future research on robust quantitative image analysis.
Journal IF-equivalent: 1.4 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Keshavarz S, Saeedzadeh E, Arabi H, Jenabi-Haghparast E, Sardari D. Self-supervised multimodal swin UNETR for PET/CT segmentation of diffuse large B-cell lymphoma. Cancer Treat Res Commun. 2026;49:101369. doi:10.1016/j.ctarc.2026.101369. PMID: 42612285.Checked: Abstract only - Abstract
Accurate differentiation between Nasopharyngeal Carcinoma (NPC) and Nasopharyngeal Lymphoma (NPL) is critical for clinical management. We aimed to develop an interpretable deep learning-radiomics fusion model using positron emission tomography/computed tomography (PET/CT) and validate it externally. In this retrospective dual-center study, we enrolled 434 patients (349 internal, 85 external) with pathologically confirmed, treatment-naive NPC or NPL who underwent baseline FDG PET/CT. We extracted 1688 radiomics features and deep learning features from 3D-ResNet50. A three-step feature selection was applied, and a multimodal logistic regression model was constructed. Model performance was evaluated on an external test set, and SHapley Additive exPlanations (SHAP) analysis was used for interpretation. In this retrospective dual-center study, a total of 434 patients (434 primary lesions) were enrolled, including 349 from the internal cohort and 85 from the external cohort. Among them, 329 patients (75.8%) had NPC ( internal: 266, external: 63) and 105 patients (24.2%) had NPL( internal: 83, external: 22). The mean age of NPC patients was 50.5 ± 13.8 years, while that of NPL patients was 59.0 ± 16.5 years. The final fusion model (denoted as the ALL model) achieved an AUC of 0.833, specificity of 95.5%, and accuracy of 75.3% in external validation, outperforming all single-modality models (all P < 0.05). SHAP analysis identified key predictors: original_shape_MinorAxisLength_PET and lbp-3D-k_glrlm_RunLengthNonUniformity_PET for NPL, and wavelet-LLL_firstorder_Maximum_CT features for NPC. We developed a robust, interpretable fusion model for NPC/NPL differentiation, offering high specificity and transparent decision logic, with potential as a non-invasive clinical decision-support tool.
Journal IF-equivalent: 3.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Liang M, Guo C, Gao X, Li H, Yang M, Jing Y, et al. A Dual-Center, Interpretable Deep Learning-Radiomics Fusion Model for Distinguishing Nasopharyngeal Carcinoma from Nasopharyngeal Lymphoma Using FDG PET/CT. Acad Radiol. 2026 Aug [Epub ahead of print]. doi:10.1016/j.acra.2026.07.060. PMID: 42603752.Checked: Abstract only - Abstract
To evaluate the predictive value of tabular deep learning models trained on handcrafted 18 F-FDG PET/CT PyRadiomics features and clinical data for cancer-related cognitive impairment (CRCI) among patients with diffuse large B-cell lymphoma without central nervous system involvement (non-CNS DLBCL). We retrospectively analyzed 112 non-CNS DLBCL patients who underwent brain 18 F-fluorodeoxyglucose ( 18 F-FDG)PET/CT. Based on Montreal Cognitive Assessment (MoCA) and Functional Assessment of Cancer Therapy-Cognitive Function (FACT-Cog) scores, patients were classified into the CRCI group (n = 73) and the non-CRCI group (n = 39) and randomly partitioned into training (n=89) and test (n=23) sets at an 8:2 ratio. Five-fold cross-validation was performed within the training set for model development, and bootstrap resampling was used to estimate 95% confidence intervals for AUC. Radiomic features were extracted with PyRadiomics from automatically segmented brain regions using MOOSE120 software. These handcrafted PET/CT radiomic features and clinical variables were organized as structured tabular datasets; raw PET/CT images were not used as direct model inputs. Feature selection was performed using correlation coefficients. The performance of seven tabular deep learning models was evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA), and different input modalities (clinical, PET, CT, and combined) were compared. Shapley additive explanations (SHAP) were applied to interpret feature contributions. The AutoInt and Neural Oblivious Decision Ensembles (NODE) models showed the best predictive performance, achieving training areas under the curve (AUCs) of 0.905 and 0.904, and test AUCs of 0.858 and 0.808, respectively. DCA suggested potential clinical utility. In the modality-specific analysis, the AutoInt model integrating CT, PET, and clinical features achieved a training AUC of 0.905 and a test AUC of 0.858, representing the most favorable test-set performance among the evaluated AutoInt input modalities. A CT texture feature from the left occipital lobe emerged as the highest-ranked model contributor in SHAP analysis, with a mean absolute attribution score of approximately 0.0078. Tabular deep learning models based on handcrafted PET/CT PyRadiomics features showed preliminary potential for predicting CRCI in patients with non-CNS DLBCL. Among the evaluated models, AutoInt achieved the most favorable performance in the exploratory internal test set. Multimodal integration of PET, CT, and clinical tabular features may provide complementary information for CRCI prediction. SHAP analysis identified a left occipital CT texture feature as the top-ranked model contributor; however, this finding should be regarded as exploratory rather than a validated biomarker.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Wang J, Dong Y, Xu N, Ma X, Liu Y, Zhao Q. Predictive value of 18F-FDG PET/CT PyRadiomics features using a tabular deep learning model for cancer-related cognitive impairment in patients with DLBCL without CNS involvement. Front Oncol. 2026 Aug 12;16:1861773. doi:10.3389/fonc.2026.1861773. PMID: 42656315; PMCID: PMC13506380.Checked: Full text checked - Abstract
Background: While maintenance rituximab therapy (MT) for follicular lymphoma improves progression-free survival, individual benefit varies, creating a critical need to avoid overtreatment and its associated burdens.
Methods: We developed a double machine learning framework using a retrospective 404-patient cohort to estimate the individualized treatment effect of MT on the risk of disease progression within 24 months.
Results: Our model revealed heterogeneity in treatment benefit, successfully distinguishing patients most likely to benefit from those who are not. A retrospective simulated application identified a "low-risk, low-benefit" subgroup that might potentially forgo MT. Furthermore, aligning historical clinical decisions with our model's recommendations was associated with a lower progression rate compared to nonalignment (14.8% vs. 38.0%). Extensive sensitivity analyses confirmed the robustness of this treatment heterogeneity against potential unmeasured confounding and temporal practice shifts.
Conclusions: Supported by an updated interactive clinical decision tool, this data-driven framework provides a robust strategy to personalize MT by separating prognostic risk from predictive benefit. It serves as a hypothesis-generating template to guide future prospective risk-adapted trials and reduce unnecessary treatment in oncology.
Journal IF: unknown (could not be matched)Reference: Gao J, Liu J, Li X, Wu M, Cui R, Huang Y, et al. Personalizing Maintenance Rituximab in Follicular Lymphoma: A Machine Learning Framework for Risk–Benefit Optimization. Health Data Sci. 2026 Aug 12;6:0467. doi:10.34133/hds.0467. PMID: 42591699; PMCID: PMC13462244.Checked: Full text checked - Abstract
Intensity-modulated radiation therapy (IMRT) is widely used for lymphoma because a multileaf collimator (MLC) can sculpt dose around irregular targets. Planning, however, is time-consuming and difficult to standardize. Fluence map prediction (FMP) using deep learning (DL) could potentially be used to automate planning, but it has not been evaluated in full-body lymphoma, where prescriptions, organ-at-risk (OAR) sets, and anatomical presentations vary substantially. Here, we evaluate multiple DL configurations to determine whether they can be adapted to lymphoma and generate IMRT plans that are deliverable in a clinical treatment planning system (TPS). To investigate the feasibility and limitations of DL-based dose and fluence prediction methods in the heterogeneous setting of lymphoma. A single-institution lymphoma cohort (571 patients, 619 plans) was replanned from clinical volumetric modulated arc therapy (VMAT) to a consistent 15-field IMRT reference using the Eclipse Scripting API (ESAPI), yielding 9285 beam-level pairs. We implemented a two-stage DL pipeline that first predicts beam-wise dose from CT and planning target volume (PTV), then maps dose to fluence, and finally computes the corresponding dose in the TPS after leaf sequencing. We evaluated several architectural and training variants, from replication-focused models to configurations with a clinical loss based on dose-volume histogram (DVH) aimed at moving beyond simple plan replication. For each configuration, we analysed how these choices influenced predicted dose realism and the quality of the resulting plans. In Stage 1 (dose prediction), replication-focused DL models were numerically inferior to the IMRT and VMAT references, while DVH-driven variants yielded superior target DVHs. After fluence mapping and TPS sequencing, all configurations produced machine-deliverable IMRT plans but with lower target coverage and higher hot spots than the references; fluence was largely preserved by MLC sequencing, indicating that current limitations mainly stem from the predicted dose distributions and the learned dose-to-fluence mapping. DL-based dose and fluence prediction can generate fast, machine-deliverable IMRT plans for heterogeneous lymphoma that lie in the clinical ballpark, but they remain dosimetrically inferior to manual IMRT and VMAT, highlighting the current limitations of these methods for automated planning in this setting.
Journal IF-equivalent: 2.4 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Kutnár D, Vogelius IR, Georgiev HA, Håkansson KE, Specht L, Petersen J. End‐to‐end deep learning pipeline for automated IMRT planning in lymphoma: Feasibility and limitations. J Appl Clin Med Phys. 2026 Aug 4;27(8):e70702. doi:10.1002/acm2.70702. PMID: 42552717; PMCID: PMC13438194.Checked: Full text checked - Abstract
Accurate grading of follicular lymphoma (FL) is crucial for personalized treatment, but biopsy-based histopathology is invasive and limited by observer variability. To address these limits, we present an artificial intelligence framework for physician-guided assisted FL grading using PET-CT imaging. Our approach integrates an enhanced dual-discriminator conditional GAN (DDCGAN) featuring similarity and chrominance constraints to generate task-specific fused images with preserved metabolic-structural cues. Furthermore, a Bayesian ResNet is introduced to explicitly model predictive uncertainty, effectively resolving classification ambiguity between adjacent FL Grades I and II. Rigorous evaluation on a multi-center dataset of 837 patients, including FL and diffuse large B-cell lymphoma (DLBCL), proves that our framework delivers superior generalizability. It achieves an accuracy of 0.871, precision of 0.875, and macro-F1 of 0.816, outperforming single-modality and existing state-of-the-art fusion models. Ultimately, this task-oriented image fusion and uncertainty-aware framework offers a highly practical, non-invasive decision-support tool to support scalable clinical decision-making in hospital workflows.
Journal IF-equivalent: 3.9 (OpenAlex 2-year mean citedness, value as of 2026-10-13, retrieved 2026-10-15; not the official Clarivate IF)Reference: Qian C, He L, Jiang Q, Zhou H, Jiang Z, Teng Y, et al. Early Follicular Lymphoma Grading via PET-CT Fusion and Bayesian Deep Learning. J Imaging Inform Med. 2026 Aug 4 [Epub ahead of print]. doi:10.1007/s10278-026-02124-6. PMID: 42552270.Checked: Abstract only - Abstract
Patients with lymphoma with bone marrow infiltration have a poor prognosis, and reliable bone marrow infiltration confirmation methods are critical for clinical management. This study aimed to construct machine learning models to identify bone marrow infiltration-influencing factors and predict lymphoma bone marrow infiltration using multimodal clinical data. A retrospective study included 243 bone marrow infiltration and 541 non-bone marrow infiltration patients with lymphoma. Eight machine learning algorithms were used to build models based on clinical characteristics, laboratory tests, and fluorine-18fluorodeoxyglucose Positron emission computed tomography (PET/CT) data. Model performance was evaluated by AUC and accuracy. A sensitivity analysis was performed in the pathology-confirmed bone marrow infiltration cohort to assess potential incorporation bias. Platelet count, hemoglobin, serum lactate dehydrogenase, lymph node involvement distribution, bone infiltration characteristics, and maximum standardized uptake value (SUV max ) of sternum, SUV max of thoracic vertebra, SUV max of lumbar vertebra, SUV max of sacrum, SUV max of femur correlated with lymphoma bone marrow infiltration. The random forest (RF) model showed the best performance (AUC = 0.842, accuracy = 80.9%) in the testing set. In pathology-confirmed subgroup analysis, the RF model maintained stable performance (AUC = 0.846, accuracy = 82.5%). The RF model is a powerful tool for lymphoma bone marrow infiltration prediction (AUC = 0.774, accuracy = 80.3%) using routine clinical, laboratory and PET/CT data, providing a promising approach for bone marrow infiltration assessment, PET/CT interpretation and quantitative risk stratification.
Journal IF-equivalent: 2.1 (OpenAlex 2-year mean citedness, value as of 2026-11-02, retrieved 2026-11-03; not the official Clarivate IF)Reference: Zhang Z, Yang T, Liu D, Sa R, Guan F. Prediction of Lymphoma Bone Marrow Infiltration by Machine Learning Models Based on Fluorine-18 Fluorodeoxyglucose Positron Emission Computed Tomography. JCO Glob Oncol. 2026 Jul;12(7):e2600078. doi:10.1200/go-26-00078. PMID: 42525912.Checked: Abstract only - Abstract
Objective : This study aimed to develop a multi-modal ultrasound-based model integrating radiomics, deep learning, and clinical features to predict progression-free survival (PFS) in diffuse large B-cell lymphoma (DLBCL) with superficial lymph node involvement. Methods : A total of 281 DLBCL patients with superficial lymph node involvement treated with standard regimens were retrospectively enrolled and assigned to training and test sets at a 7:3 ratio. Ultrasound radiomic features were extracted by PyRadiomics, and deep learning features were derived from DenseNet121 with transfer learning. After Pearson's correlation filtering and least absolute shrinkage and selection operator (LASSO)-Cox regression for feature selection, five prognostic models were compared using C-index, time-dependent Area Under the Curve (AUC), risk stratification and decision curve analysis. Results : The combined model achieved the highest C-index of 0.811 in the test set, with 1-/2-/3-year PFS AUCs of 0.826, 0.812, and 0.810, respectively. Risk stratification showed significantly poorer PFS in the high-risk group ( p < 0.01). A visualized nomogram was further developed as a preliminary reference for individualized prediction. Conclusions : This multi-modal combined predictive model and corresponding nomogram based on superficial lymph node features showed promising potential for practical and intuitive prognostic assessment and risk stratification in DLBCL.
Journal IF-equivalent: 4.0 (OpenAlex 2-year mean citedness, value as of 2026-10-29, retrieved 2026-10-31; not the official Clarivate IF)Reference: Wang J, Mu J, Yang Y, Meng X, Song Y. Multi-Modal Ultrasound-Based Prognostic Model for Diffuse Large B-Cell Lymphoma with Predominantly Superficial Lymph Node Involvement. Diagnostics (Basel). 2026 Jul 23;16(15):2311. doi:10.3390/diagnostics16152311. PMID: 42587550; PMCID: PMC13465404.Checked: Full text checked - Abstract
To develop a PET/CT-based radiomics model for noninvasive prediction of the double-expressor lymphoma (DEL) subtype in diffuse large B-cell lymphoma (DLBCL). From January 2019 to July 2025 consecutively enrolled, 143 patients with DLBCL (55 DEL, 88 non-DEL) were randomly divided into a training set and an internal validation set in a 7:3 ratio. Radiomic features were extracted from peritumoral regions with different expansion distances (3, 5, and 10 mm) to identify the optimal peritumoral region. These features were then integrated using various fusion strategies (multi-region fusion, image fusion, and feature fusion) for combined intratumoral and peritumoral analysis. A transfer learning model built on a pretrained ResNet-50 was combined with a clinical model incorporating baseline PET features to develop three integrated models (Clinic+Rad+DL, Stacking, Ensemble). Model performance was evaluated to select the best-performing approach, and SHAP analysis was applied to enhance interpretability. The Rad_MLP model achieved the best performance by integrating intratumoral and 10-mm peritumoral features, with an accuracy of 85.3%, a sensitivity of 89.5%, and a specificity of 79.2%. The positive predictive value (PPV) and negative predictive value (NPV) were 0.773 and 0.905, respectively, with an F1 score of 0.829. Rad_MLP, by integrating the most predictive intratumoral and peritumoral features, substantially improves the accuracy of noninvasive prediction of the DEL subtype in DLBCL.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Zhong Q, Lu G, Liang J, Peng Q, Tian R, Zhang G. Optimal peritumoral regions and fusion strategies for prediction of the double-expressor subtype in diffuse large B-cell lymphoma: a multi-region radiomics study. Front Oncol. 2026 Jul 23;16:1877578. doi:10.3389/fonc.2026.1877578. PMID: 42564139; PMCID: PMC13442022.Checked: Full text checked - Abstract
The presence of diffuse large B-cell lymphoma (DLBCL) cells in bone marrow (BM) smears is recognised as a key morphological basis for diagnosing BM involvement by DLBCL. BM involvement directly affects disease staging, treatment strategies, and prognosis assessment. Nevertheless, traditional manual identification of DLBCL cells in BM smears under a microscope is time-consuming, subject to observer variability, and lacks standardization between hospitals. In this study, we developed an artificial intelligence (AI) detection system (DLBCL-CQnet) based on deep learning, using BM smears from 117 patients with newly diagnosed DLBCL across four centers. DLBCL-CQnet comprises a region of interest (ROI) classification model based on MobileNetV2, and a cell detection and classification model (CDC-MOD) based on YOLOv8. Performance was evaluated using accuracy, precision, recall, F1-score, and mean average precision at an intersection over union threshold of 0.5 (mAP50). MobileNetV2 achieved the best ROI classification performance (accuracy 0.925, precision 0.922, F1-score 0.919), while YOLOv8 yielded a classification accuracy of 0.890 and an mAP50 of 0.934 for DLBCL cell detection. The system achieved an ROI classification accuracy of 0.925, a cell classification accuracy of 0.890, and a mean average precision (mAP50) of 0.934 at an intersection-over-union (IoU) threshold of 0.5. The optimal cutoff for identifying BM infiltration by DLBCL cells was 2.12%, with a specificity of 0.941 and positive predictive value (PPV) of 0.923, demonstrating high diagnostic value. The recognition speed significantly outperforms manual assessment, substantially reducing time consumption. DLBCL-CQnet enables automated, standardized identification of DLBCL cells in BM smears across institutions, serving as a reliable auxiliary diagnostic tool for assessing BM involvement by DLBCL. Its high specificity and operational efficiency streamline diagnostic workflows, especially in resource-constrained environments.
Journal IF-equivalent: 5.9 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Ran L, Peng Y, Wu X, Yao Y, Li Z, Chen S, et al. Development and multicenter validation of an AI-driven detection system for DLBCL cells in bone marrow smears. Sci Rep. 2026 Jul 17;16(1):32775. doi:10.1038/s41598-026-59908-7. PMID: 42469316.Checked: Abstract only - Abstract
The clinical course of Mantle Cell Lymphoma (MCL) varies between individual patients. Early detection of risk is crucial to assign MCL patients to novel treatment strategies. Most of the established biomarkers of outcome require specifically trained pathologists or molecular analysis. Here we introduce MAIPI (MCL Artificial Intelligence Prognostic Index), a deep learning algorithm trained only on Hematoxylin and Eosin (H&E) images of diagnostic biopsies of n = 428 MCL patients from clinical trials to assess prognosis. The capability of MAIPI to predict disease outcome was validated in an independent cohort of n = 140 patients treated with immunochemotherapy with and without ibrutinib. MAIPI selects areas of interest by itself and provides prognostic information independent of the MCL International Prognostic Index (MIPI) and Ki67 and without the need of molecular testing or expert pathologists evaluation.
Journal IF-equivalent: 7.4 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Lippl J, Reinke S, Koch K, Oschlies I, Schrod S, Wolfseher L, et al. Mantle cell lymphoma artificial intelligence prognostic index using hematoxylin and eosin histology. Leukemia. 2026 Jul 7;40(9):1961-7. doi:10.1038/s41375-026-03036-8. PMID: 42414604; PMCID: PMC13506322.Checked: Full text checked - Abstract
Diffuse large B-cell lymphoma (DLBCL) is characterized by profound heterogeneity that underpins varied clinical outcomes. To decipher this complexity, we performed an integrated single-cell and genomic analysis. Using scRNA-seq data (GSE182434), we identified six distinct malignant B-cell subclusters (MB1-MB6) within the DLBCL ecosystem. Cell-cell communication analysis revealed intricate interaction networks, particularly involving the MIF and Complement pathways. Prognostic analysis of bulk transcriptomic data (GSE32918) identified the MB5-related gene signature as the most critical factor associated with poor overall survival. This MB5 subgroup was associated with enhanced proliferative processes, a higher tumor mutational burden, and specific comutations. Leveraging MB5 marker genes, we developed and validated a robust CoxBoost-RSF machine-learning model that effectively stratified patient risk in independent cohorts. Our study defines the MB5 malignant B-cell subgroup as a key driver of DLBCL aggressiveness and provides both a novel prognostic biomarker and a framework for personalized therapeutic targeting.
Journal IF-equivalent: 1.0 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Shen H, Lu J, Yan Q, Chou J, Liang X, Fan W, et al. Identifying Distinct Molecular Subtypes and Establishing a Prognostic Framework for DLBCL Patients via Multiomics Analysis and Machine Learning Approaches. Hum Mutat. 2026 Jul 6;2026(1):7614954. doi:10.1155/humu/7614954. PMID: 42440855; PMCID: PMC13334209.Checked: Full text checked - Abstract
Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous malignancy driven by complex immune dysregulation, yet the role of NETosis, a distinct neutrophil immune effector program, in the tumor immune microenvironment and patient prognosis remains poorly defined. This study leveraged an integrative multi-omics and machine learning framework to elucidate the immunological implications of NETosis-related genes (NETosis-RGs) in DLBCL. NETosis-RGs were identified from transcriptomic profiles of DLBCL versus control tissues (GSE32018). Prognostic candidates were screened by univariate Cox regression and modeled using five machine learning algorithms, with the Random Survival Forest (RSF) model selected for risk stratification. Bulk and single-cell transcriptomic data were integrated to resolve the immune landscape, while ceRNA networks and m 6 A regulator analyses were performed to characterize post-transcriptional and epigenetic layers. Chemotherapy sensitivity was predicted to explore therapeutic vulnerabilities. The RSF-based model, incorporating five key NETosis-RGs (AKT1, SLC25A37, FPR2, TLR7, F3), robustly stratified patients into high- and low-risk groups exhibiting systematically distinct immune architectures-spanning differential immune cell infiltration, stromal/immune scores, and immune checkpoint expression. Single-cell transcriptomic analysis further revealed that these prognostic genes were predominantly expressed in monocytes and M1 macrophages and displayed dynamic regulation along myeloid/macrophage differentiation trajectories, suggesting their role as potential modulators of the immunosuppressive-to-inflammatory transition within the DLBCL microenvironment. Complementary analyses revealed multi-layered regulatory mechanisms, including m 6 A modification patterns and lncRNA-miRNA-mRNA ceRNA axes. By integrating multi-omics data with machine learning, this study delineates a novel NETosis-centered immune signature that refines prognostic stratification and provides mechanistic insights into monocyte/macrophage-mediated immune dysregulation in DLBCL. The findings offer a framework for future immunomodulatory biomarker development and the development of precision therapeutic strategies.
Journal IF-equivalent: 4.8 (OpenAlex 2-year mean citedness, value as of 2026-10-15, retrieved 2026-10-16; not the official Clarivate IF)Reference: Yao H, Wang G, Liang Y, Zhang R. Decoding NETosis-associated immune dysregulation in diffuse large B-cell lymphoma through integrative multi-omics and machine learning. Front Cell Dev Biol. 2026 Jul 2;14:1867660. doi:10.3389/fcell.2026.1867660. PMID: 42465039; PMCID: PMC13372773.Checked: Full text checked - Abstract
Primary central nervous system lymphoma (PCNSL) exhibits substantial biological heterogeneity, particularly across immunohistochemical subtypes such as double-expression lymphoma (DEL) and germinal center B-cell-like (GCB) phenotypes. This study aimed to develop and evaluate multiparametric MRI-based radiomics models for preoperative prediction of DEL andGCB status in PCNSL. We retrospectively included 160 pathologically confirmed PCNSL patients. Multiparametric MRI sequences, including T2-weighted (T2WI), T2-Fluid-Attenuated Inversion Recovery (FLAIR), contrast-enhanced T1-weighted (T1CE), and apparent diffusion coefficient (ADC), were analyzed. Enhancing tumor core and peritumoral edema were automatically segmented using nnU-NetV2-based models, and radiomics features were extracted from both regions across all sequences. After reproducibility filtering, ComBat harmonization, and multistep feature selection performed exclusively within the training cohort, six machine-learning classifiers were trained and evaluated in held-out internal test sets. Model performance was assessed using ROC and decision curve analysis, and SHAP-based feature interpretation. For DEL classification, 12 radiomic features were retained. The SVM classifier achieved the best test performance, with an area under the ROC curve (AUC) of 0.807 (95% CI, 0.649-0.936), accuracy of 0.730, sensitivity of 0.714, and specificity of 0.750. For GCB/non-GCB classification, seven radiomic features were used for model construction. The Random Forest classifier achieved the highest test performance, with an AUC of 0.897 (95% CI, 0.796-0.973), accuracy of 0.846, sensitivity of 0.792, and specificity of 0.893. Multiparametric MRI-based radiomics analysis demonstrated promising performance for noninvasive prediction of DEL and GCB/non-GCB phenotypes in PCNSL. These findings suggest that MRI-derived radiomic features may capture imaging correlates of biological heterogeneity and may support preoperative risk stratification and individualized treatment planning.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Chen L, Wang X, Wang S, Chen T, Zhao X, Yan Y, et al. Preoperative phenotypic stratification of primary central nervous system lymphoma using multiparametric MRI-based radiomics: prediction of germinal center B-cell-like and double-expression status. Front Oncol. 2026 Jun 29;16:1822040. doi:10.3389/fonc.2026.1822040. PMID: 42444831; PMCID: PMC13357172.Checked: Full text checked - Abstract
This study aims to create and validate a prognostic prediction model that leverages deep features derived from PET/CT imaging to support personalized precision treatment for patients with diffuse large B-cell lymphoma (DLBCL). We retrospectively analyzed clinical and pretreatment PET/CT data from 209 patients with DLBCL. Deep features were extracted from three-dimensional tumor lesions; after dimensionality reduction, a radiomics model was built to predict 3-year overall survival (OS). We evaluated six machine learning algorithms: Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM). The optimal radiomics model was integrated with clinical features to form a fusion model. The fusion model's performance was assessed on both training and independent test sets using metrics such as accuracy, area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, and specificity. Furthermore, the model's clinical utility was evaluated through Decision Curve Analysis (DCA), and survival analysis was performed using the Kaplan-Meier (KM) method. Univariate and multivariate analyses identified age, AB group, International Prognostic Index (IPI) score, serum β2-microglobulin level, and maximum tumor diameter as independent risk factors for 3-year survival in DLBCL patients. Among the machine learning-based radiomics models, the LR model showed superior predictive performance, achieving an accuracy of 0.865, AUC of 0.950, sensitivity of 0.875, and specificity of 0.863. Integration with clinical features further improved the model's performance. On the test set, the fusion model attained an accuracy of 0.921, an impressive AUC of 0.974, and sensitivities and specificities of 0.846 and 0.940, respectively. DCA revealed that this fusion model offers a significant clinical net benefit for prognostic risk prediction in DLBCL patients over a wide threshold probability range of 0.05-0.900. The fusion model, combining PET/CT deep features with clinical characteristics, is a dependable prognostic tool for DLBCL, holding substantial promise for clinical application and personalized treatment.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Wang M, Wu S, Cen Q, Yang H, Zhou S, Qiu J, et al. A deep learning-based prognostic model for diffuse large B-cell lymphoma incorporating PET/CT imaging features. Front Oncol. 2026 Jun 16;16:1849942. doi:10.3389/fonc.2026.1849942. PMID: 42382391; PMCID: PMC13314506.Checked: Full text checked - Abstract
Accurate prediction of treatment response in Hodgkin lymphoma (HL) is crucial for personalized therapy. The Tensor Radiomics (TR) paradigm advances traditional radiomics by producing and analyzing diverse feature variations, employing tensors computed across multiple parameter combinations to optimize radiomics feature extraction and improve tumor characterization. However, the complexity and variability of these features pose significant challenges for analysis and clinical application in HL research. This study introduces the Tensor Radiomics Network (TR-NET), a deep learning framework that automates feature selection from tensor radiomics data and predicts treatment response. It also incorporates explainable AI techniques to identify the most influential features. We analyzed 420 lesions from 70 HL patients using 18 F-FDG PET/CT. Tensor radiomic feature sets were generated by varying bin widths for PET (0.2, 0.4, 0.6, 0.8) and CT (10, 25, 50) images. Extracted features encompassed first-order, shape, texture, and filter classes (Laplacian of Gaussian, Wavelet, Square, Square Root, Logarithm, Exponential, Gradient, Local Binary Pattern (LBP) 2D, LBP 3D for each lesion, totaling 13,790 features. We utilized machine learning models, including Random Forest (RF), XGBoost, Support Vector Machine (SVM), and the deep learning model TR-NET for analysis. SHAP (SHapley Additive exPlanations), an explainable AI method, was applied to interpret the predictions of the best-performing model. TR-NET achieved a predictive performance, with an AUC-ROC of 0.8470 (95% CI: 0.7669-0.9185), a sensitivity of 82.2%, and a specificity of 78.2%, offering a balanced classification characteristic. In comparison, XGBoost achieved an AUC of 0.8160 (95% CI: 0.7457-0.8599), characterized by high sensitivity (93.3%) but lower specificity (66.2%), indicating a tendency towards false positives. Random Forest and SVM also showed high sensitivity (92% and 91.1%), but both suffered from significantly reduced specificity (29.5% and 51.3%). TR-NET demonstrates enhanced predictive accuracy and clinical relevance compared to classic machine learning, supporting personalized treatment strategies and the integration of explainable AI in oncology.
Journal IF-equivalent: 3.7 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Jajroudi M, Jamalirad H, Enferadi M, Roshanravan V, Emami F, Geramifar P, et al. A deep learning framework for lesion-level treatment response prediction in hodgkin lymphoma using PET/CT tensor radiomics. EJNMMI Phys. 2026 Jun 15;13(1):64. doi:10.1186/s40658-026-00838-8. PMID: 42295572; PMCID: PMC13269603.Checked: Full text checked - Abstract
Cytomegalovirus (CMV) reactivation occurs in 30 - 40 % of seropositive patients receiving chimeric antigen receptor T-cell (CAR-T) therapy for B-cell lymphoma and is strongly associated with treatment failure. However, the causal immunological mechanism driving this failure whether through direct viral cytopathic effects, T-cell exhaustion, or resource competition remains undefined. Furthermore, existing predictive models lack the mechanistic insight needed to guide intervention. We developed a privacy-preserving, mechanistic digital twin to test the "cytokine sink" hypothesis, wherein CMV-specific CD8+ T cells compete with CAR-T cells for the limiting homeostatic cytokine I L - 15 . We constructed a system of ordinary differential equations formalizing this competition for I L - 15 . The model was trained using a Hierarchical Bayesian Federated Averaging algorithm on multi-institutional data from 414 patients across five academic cancer centres (Sites A-E). without centralizing raw patient data. An independent sixth centre (Site F, n = 89 ) served exclusively as the prospective validation site and contributed no data to the training process. In silico counterfactual analyses and simulations of antiviral prophylaxis were performed. The digital twin predicted clinically significant CMV reactivation defined as CMV viral load ≥ 10 , 000 I U / m L by D a y 28 post-infusion (primary outcome; threshold selected based on E C I L - 7 guidance for threshold-triggered pre-emptive antiviral therapy in T-cell-depleted immunotherapy) with an AUROC of 0.91 ( 95 % C I : 0.88 - 0.94 ) , significantly outperforming existing clinical risk scores. Two pre-specified secondary thresholds were also analysed: (i) any detectable reactivation ( ≥ 1 , 000 I U / m L ) and (ii) severe reactivation ( ≥ 50 , 000 I U / m L ). Under the structural assumptions of the fitted mechanistic model, the data are quantitatively consistent with I L - 15 resource competition as a mechanism linking CMV reactivation to CAR-T impairment. Mechanistically, the model revealed that CMV reactivation was associated with reduced peak CAR-T expansion by 41.8 % ( p < 0.001 ) in counterfactual analysis. Global sensitivity analysis identified the pre-infusion frequency of CMV-specific T-cell precursors ( ξ ) and the resource competition coefficient ( α ) as the primary drivers of this effect, explaining 38 % and 29 % of output variance, respectively. In silico simulation of a risk-adapted, digital-twin-guided antiviral prophylaxis strategy reduced projected six-month progression by 32 % while reducing aggregate drug exposure by 36 % . In the prospective validation cohort, the model-predicted kinetic impairment independently predicted progression-free survival (hazard ratio [HR] 3.4 , 95 % C I : 1.5 - 7.8 , p = 0.004 ) . Formal model competition analysis against three alternative mechanistic hypotheses (Results Section 3.3) further supports the cytokine sink hypothesis over exhaustion-based or mediation-only alternatives. Under the structural assumptions of the fitted mechanistic model, the data are quantitatively consistent with I L - 15 resource competition as a mechanism linking CMV reactivation to CAR-T impairment. Formal model competition analysis against three alternative mechanistic hypotheses further supports the cytokine sink hypothesis; randomised interventional evidence is required for definitive causal proof. We further demonstrate that a privacy-preserving, mechanistic digital twin can serve as a clinically actionable tool for early risk stratification and personalized intervention, while providing a scalable blueprint for collaborative systems immunology research. Definitive validation of clinical utility requires a randomised controlled trial in which treatment decisions are prospectively guided by the digital twin's predictions.
Journal IF-equivalent: 6.0 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Sridharan P, Ghosh M. A federated digital twin reveals cytomegalovirus reactivation impairs CAR-T cell therapy via IL-15-mediated cytokine competition in B-Cell lymphoma. Front Immunol. 2026 Jun 12;17:1844927. doi:10.3389/fimmu.2026.1844927. PMID: 42367785; PMCID: PMC13307505.Checked: Full text checked - Abstract
Peripheral T-cell lymphoma-not otherwise specified (PTCL-NOS) is a highly aggressive and heterogeneous lymphoma subtype with a poor prognosis. This study aims to develop a machine learning-based model to predict early death (within 3 months of diagnosis) in PTCL-NOS patients using data from the SEER database (2016-2021). A total of 1,156 patients were included and randomly divided into training (n = 809) and validation (n = 347) sets. Key predictive factors were identified through Boruta and LASSO algorithms, including chemotherapy, radiotherapy, age, B symptoms, primary tumor site, Summary Stage, and Ann Arbor Stage. Seven machine learning models were constructed and evaluated using AUROC, AUPRC, calibration curves, Brier scores, and decision curve analysis. XGBoost demonstrated the best predictive performance (AUROC = 0.842 in training and 0.774 in validation). This study provides a novel and interpretable predictive tool that can aid in early risk stratification and personalized treatment planning for PTCL-NOS patients, ultimately improving clinical outcomes.
Journal IF-equivalent: 2.4 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Liang C, Zhong R, Li L, Zhang C, Hu L, Chen C, et al. Prediction of early death in peripheral T-cell lymphoma-NOS patients based on machine learning. Ann Hematol. 2026 Jun 10;105(9):370. doi:10.1007/s00277-026-07121-x. PMID: 42268418; PMCID: PMC13473289.Checked: Full text checked - Abstract
To clarify the cellular origin of matrix metalloproteinase-9 (MMP9) and explore targeted research, we utilized single-cell RNA sequencing analysis, which revealed that MMP9 is predominantly enriched in specific macrophages within the activated B-cell-like (ABC) subtype. Guided by this target information, we applied a generative AI pipeline incorporating RFdiffusion, ProteinMPNN, and AlphaFold to de novo design protein binders targeting the hemopexin (PEX) domain of MMP9. ELISA experiments confirmed the in vitro binding capability of these designs; among them, MMP9-30 displayed the strongest binding, with an apparent EC50 of approximately 1.1 μM, followed by MMP9-34, while MMP9-97 showed the weakest interaction. This study successfully integrates single-cell sequencing with AI-assisted protein design, providing a preliminary exploratory framework for subsequent MMP9-targeted research and protein binder development.
Journal IF-equivalent: 7.0 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Miao Z, Zhu S, Qin L, Ma D, Lai M, Xu P, et al. Integrating Single-Cell Profiling with Generative AI for De Novo Design of MMP9 Protein Binders in Diffuse Large B-Cell Lymphoma. Molecules. 2026 Jun 5;31(11):1969. doi:10.3390/molecules31111969. PMID: 42280270; PMCID: PMC13257871.Checked: Full text checked - Abstract
Diffuse large B-cell lymphoma (DLBCL) is the most common aggressive non-Hodgkin lymphoma and is characterized by substantial heterogeneity. This study aimed to develop a liquid-liquid phase separation (LLPS)-related prognostic model to improve risk stratification. Transcriptomic and clinical data from four cohorts (n = 768) were analysed. Multiple machine learning algorithms were applied to identify prognostic LLPS-related genes (LRGs) and construct a 6-LRG model. Model performance was assessed using survival analysis, time-dependent receiver operating characteristic curves and multivariable modelling. Additional analyses were conducted to explore potential biological and microenvironmental differences between risk groups. The 6-LRG model stratified patients into groups with significantly different overall survival across datasets, with 1-year area under curve (AUCs) ranging from 0.661 to 0.820, 3-year AUCs from 0.683 to 0.779 and 5-year AUCs from 0.711 to 0.807. The 6-LRG model remained independent of established clinical variables and improved risk prediction when integrated into a nomogram. Distinct biological and immune characteristics were observed between groups. The 6-LRG model may provide additional prognostic information in DLBCL and generates hypotheses regarding underlying biological mechanisms. However, prospective validation in larger populations is essential before any implementation.
Journal IF-equivalent: 3.0 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Zhou Z, Lu J, Wang Z, Chen R, Li H, Chen W, et al. Machine learning‐driven investigation on liquid–liquid phase separation‐related prognostic signature in diffuse large B‐cell lymphoma. Br J Haematol. 2026 Jun 5;209(2):461-8. doi:10.1111/bjh.70601. PMID: 42246608; PMCID: PMC13461964.Checked: Full text checked - Abstract
To develop baseline ¹ 8 F-FDG PET/CT habitat radiomics and dual-channel deep learning models for the prediction of early metabolic response (EMR) on interim PET (iPET) in patients with DLBCL and compare their performance. Patients with DLBCL who underwent baseline 18 F-FDG PET/CT between December 2018 and August 2024 and received 2-4 cycles of R-CHOP or R-CHOP-like chemotherapy were retrospectively enrolled. Based on iPET Deauville scores, patients with scores of 1-3 were classified as EMR, whereas those with scores of 4-5 were classified as non-EMR. Lesions were semi-automatically segmented to generate volumes of interest (VOIs). Voxel-wise habitat subregions were delineated using K-means clustering to characterize intratumoral heterogeneity. Radiomics features were extracted from whole-tumor and habitat subregions on PET and CT images. For deep learning, the largest axial tumor slice from PET and CT was concatenated along the channel dimension as dual-channel input. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, balanced accuracy, sensitivity, specificity, and F1 score. Calibration curves, decision curve analysis, net reclassification improvement, and integrated discrimination improvement were additionally used for model comparison. A total of 148 patients (101 EMR, 47 non-EMR) were randomly split into training (n = 103) and test (n = 45) sets. This retrospective single-center study used the training set for feature selection and model development, and the test set exclusively for final evaluation. We developed habitat radiomics and dual-channel deep learning models; multilayer perceptron and DenseNet-161 were selected as the optimal architectures, respectively. In the test set, the habitat radiomics model (Habitat_MLP) achieved an AUC of 0.871 (95% CI: 0.7563-0.9857), with a specificity of 0.903, and accuracy of 0.822, showing higher overall performance than the dual-channel deep learning model (DL_DenseNet161), which achieved an AUC of 0.793 (95% CI: 0.6409-0.9444), specificity of 0.677, and accuracy of 0.711. Furthermore, the Habitat_MLP model showed more favorable calibration, higher net benefit on DCA, and improved risk reclassification in this dataset. The habitat radiomics model derived from baseline 18 F-FDG PET/CT (Habitat_MLP) demonstrated superior overall performance and robustness for predicting EMR on iPET in patients with DLBCL, suggesting its potential value as a decision-support tool for pretreatment risk stratification.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: He Y, Wang S, Li Y, Li X, Yi J, Wang D, et al. Baseline 18F-FDG PET/CT habitat radiomics versus dual-channel deep learning for predicting interim PET early metabolic response in diffuse large B-cell lymphoma: a comparative study. Front Oncol. 2026 Jun 4;16:1801204. doi:10.3389/fonc.2026.1801204. PMID: 42326344; PMCID: PMC13275354.Checked: Full text checked - Abstract
SERPING1, which encodes the C1 inhibitor (C1-INH) of the complement system, and plays a key regulator in regulating inflammatory responses and immune homeostasis. SERPING1 is downregulated in various disease, this downregulation occurs through the body's negative feedback resulting from the overactivation of the complement system in diseases such as infections and acute inflammatory responses. Additionally, SERPING1 is vital for tumor immunomodulation. Diffuse large B-cell lymphoma (DLBCL) is a common and aggressive type of non-Hodgkin lymphoma. More treatment options are becoming available for this disease. However, some patients still experience recurrence or even disease progression during treatment. Consequently, elucidating the molecular underpinnings of DLBCL's malignant behavior and identifying novel prognostic markers and therapeutic targets are paramount for improving patient outcomes. This investigation utilized multi-omics integration combined with machine learning algorithms to identify CD8 + T cell-associated hub genes and clarify their roles in DLBCL pathogenesis. Data were obtained from public repositories including the Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA). Utilizing multi-omics and machine learning techniques such as differential analysis, WGCNA, feature selection, scRNA-seq analysis, molecular docking, and Mendelian randomization, we identified four key hub genes (MAFB, TMEM176A, SERPING1, and C1QB) linked to CD8 + T cells, highlighting SERPING1 as central. They have potential value in the diagnosis, prognosis, and immunotherapy of DLBCL. Analysis of six GEO datasets (GSE56315, GSE25638, GSE12453, GSE12195, GSE32018, GSE83632) and TCGA-DLBC RNA-sequencing data, following batch correction, revealed 209 differentially expressed genes (DEGs). CIBERSORT revealed significant immune infiltration disparities, with CD8 + T cells notably enriched in DLBCL. WGCNA identified the yellow module (151 genes) as strongly correlated with CD8 + T cell infiltration, yielding 40 core DEGs upon intersection with DEGs. Functional enrichment analysis highlighted their involvement in chemokine signaling and complement cascades. A machine learning framework utilizing 12 algorithms identified eight key genes: ANKRD22, C1QB, C1QC, CD163, MAFB, SERPING1, TMEM176A, and TNFSF13B. The combination of Least Absolute Shrinkage and Selection Operator (LASSO) with Linear Discriminant Analysis (LDA) demonstrated the highest diagnostic efficacy, achieving an area under the curve (AUC) of 0.799. Four hub genes (MAFB, TMEM176A, SERPING1, and C1QB) were identified from the eight key genes through the intersection of eight feature selection methods and validated using protein-protein interaction networks and receiver operating characteristic curve analysis, achieving an AUC greater than 0.9. SHapley Additive exPlanations analysis underscored the predictive dominance of MAFB, while a prognostic nomogram incorporating these genes demonstrated high accuracy (1-/3-/5-year calibration). scRNA-sequencing revealed hub gene enrichment in non-classical monocytes, and CellChat implicated their roles in intercellular communication via the APP/GALECTIN pathways. Molecular docking and dynamics confirmed the stable binding of SERPING1-doxorubicin (ΔG = - 7.6 kcal/mol), supported by the root mean square deviation/Rg stability. Summary-data-based Mendelian randomization analysis identified a closed positive genetic association between SERPING1 overexpression and DLBCL incidence risk (P < 0.05), with no statistical evidence of horizontal pleiotropy, suggesting that SERPING1 overexpression may be a potential risk-related factor for DLBCL onset. SERPING1 is significantly upregulated in DLBCL tissues, suggesting that its expression is regulated in a disease-specific manner. SERPING1 exhibits a pattern characterized by "local selective activation and overall inhibition" through the unbalanced regulation of the complement system in the tumor microenvironment, acting as a pro-tumor hub gene in synergistic regulation with C1QB.
Journal IF-equivalent: 2.4 (OpenAlex 2-year mean citedness, value as of 2026-10-12, retrieved 2026-10-13; not the official Clarivate IF)Reference: Wang J, Li Z, Zhan X, Zhang Y, Wang Z, Xing P, et al. An integrated machine learning and mendelian randomization approach identifies SERPING1 as a prognostic biomarker associated with CD8 + T-cell infiltration in DLBCL. Discov Oncol. 2026 Jun 3;17(1):1134. doi:10.1007/s12672-026-05340-2. PMID: 42236612; PMCID: PMC13447665.Checked: Full text checked - Abstract
The aim of this study was to evaluate the performance of an artificial intelligence (AI)-based method for automated segmentation of total metabolic tumor volume (TMTV) in 18 F-FDG PET/CT scans of patients with lymphoma, using an independent, publicly available benchmark dataset curated and segmented by expert readers in a previously published study.
Methods: The AI model, based on a 3-dimensional U-Net architecture implemented in MONAI (the medical open-source network for AI framework), was trained on 1,500 18 F-FDG PET/CT scans of patients with lymphoma. It was tested on a benchmark dataset comprising 60 baseline scans (20 each of follicular lymphoma, Hodgkin lymphoma, and diffuse large B-cell lymphoma), each segmented by 3 or 4 nuclear medicine physicians using an SUV threshold of 4. Agreement between AI-derived and benchmark TMTVs was assessed using Bland-Altman analysis, with acceptable deviation defined as within 10% or 10 cm 3 , consistent with interreader variability reported in the benchmark study.
Results: In 50 (83%) of the 60 benchmark cases, AI-derived TMTVs were within 10% or 10 cm 3 of the benchmark reference. In 4 of the remaining 10 cases, AI-derived results were within the same margin of at least 1 of the expert readers, indicating partial concordance.
Conclusion: The AI-based method achieved high concordance with expert-derived TMTVs in a standardized benchmark setting. The findings demonstrate that the AI model performs comparably to human experts in most cases, even in an externally curated dataset deliberately enriched with challenging cases by its original authors. The AI model's ability to produce accurate, reproducible segmentations without user interaction could significantly reduce manual workload and interreader variability in lymphoma imaging. However, human supervision is required to minimize errors.
Journal IF-equivalent: 6.9 (OpenAlex 2-year mean citedness, value as of 2026-10-17, retrieved 2026-10-18; not the official Clarivate IF)Reference: Sadik M, Larsson M, Enqvist O, Edenbrandt L, Trgrdh E. Validation of an AI Method for Automated Lymphoma Metabolic Tumor Volume Segmentation Using a Public Benchmark PET/CT Dataset. J Nucl Med. 2026 Mar 5;67(6):1001-5. doi:10.2967/jnumed.125.271605. PMID: 41786479; PMCID: PMC13224908.Checked: Full text checked - Abstract
Accurate lesion segmentation on 18F-FDG PET/CT is essential for the effective management of Diffuse Large B-cell Lymphoma (DLBCL). While deep learning architectures, particularly U-Net and its variants, are widely used, identifying the most reliable and robust framework for clinical quantification remains a challenge. This study aims to perform a systematic, three-way comparative evaluation of representative deep learning approaches to establish a performance hierarchy and clarify the source of superiority. A systematic comparison was performed among three architectures: a manually configured classic U-Net, an advanced Attention U-Net (incorporating attention gates), and the self-configuring nnU-Net framework, using 217 annotated whole-body PET/CT scans. The classic and Attention U-Nets were explicitly designed and tuned by experts, whereas nnU-Net automated its entire pipeline. All models were evaluated via 3-fold cross-validation using the Dice Similarity Coefficient (DSC), Intersection over Union (IoU), sensitivity, and specificity. Clinical relevance was assessed by comparing model-derived Total Metabolic Tumor Volume (TMTV) with expert annotations using correlation and Bland-Altman analyses. The final ground-truth segmentation for each patient was generated through a consensus process between the two expert annotators. The results consistently favored nnU-Net across all evaluated metrics, supporting a robust performance hierarchy among the evaluated models. It achieved a higher mean DSC (0.83 vs. 0.61 vs. 0.71), IoU (0.71 vs. 0.52 vs. 0.62), sensitivity (0.85 vs. 0.56 vs. 0.72), and specificity (0.94 vs. 0.87 vs. 0.92). nnU-Net further showed improved stability, reflected by lower performance variability across cross-validation folds. In TMTV estimation, nnU-Net showed substantially reduced systematic error and much narrower Bland-Altman limits of agreement (-15.8 to +10.5 mL vs. -108.5 to +71.1 mL vs. -97.3 to +67.7). These findings suggest that the observed advantages of nnU-Net are primarily associated with comprehensive pipeline optimization rather than architectural modifications alone. This comparative analysis demonstrates that the self-configuring nnU-Net framework provides superior and more stable segmentation performance compared with both manually designed U-Net variants, enabling more accurate and reliable DLBCL lesion delineation and TMTV quantification. Its automated pipeline and robust performance support its integration into clinical workflows. Future studies will focus on multi-center external validation, correlation with clinical outcomes (e.g., survival), and model interpretability.
Journal IF-equivalent: 2.2 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Keshavarz S, Saeedzadeh E, Arabi H, Jenabi-Haghparast E, Sardari D. Clinically reliable and stable automated segmentation of DLBCL lesions on PET/CT using self-configuring nnU-Net for robust TMTV quantification. Appl Radiat Isot. 2026 Sep;235:112721. doi:10.1016/j.apradiso.2026.112721. PMID: 42214823.Checked: Abstract only - Abstract
Pediatric lymphoma patients undergo multiple 18 F-FDG PET/CT examinations for staging and response assessment, raising concerns about cumulative radiation dose, particularly from the CT component. We propose SnapPET, a CT-sparing deep learning-based framework that uses 2D ultra-short non-attenuation-corrected (NAC) PET (Maximum intensity projection) MIP images to generate 2D high-quality PET MIP images for rapid screening and triage. Two hundred pediatric patients with FDG-avid lymphoma imaged on a total-body PET/CT scanner were retrospectively analyzed. From 10-min list-mode data, non-attenuation-corrected (NAC) PET images at 3-300 s were retrospectively reconstructed, and 600-s attenuation-corrected (AC) PET served as the reference. Whole-body coronal maximum-intensity projections (MIPs) were converted into 2D images and divided into 64 × 64 patches. For each acquisition duration, an independent latent diffusion-based autoencoder was trained to map the corresponding short-duration NAC PET MIPs to AC-like coronal PET MIPs. The intended clinical application was rapid pre-CT triage, with primary emphasis on the ultra-short 3-s setting, while longer durations were analyzed comparatively across count levels. Image quality was evaluated using PSNR, SSIM, and NMSE, as well as lesion ROI-based SUVmax and SUVmean. Three benchmark models (U-Net, GAN, and DDPM) were trained under identical settings for comparison. In addition, two nuclear medicine physicians assessed image quality using 5-point Likert scales, and lesion-level ROC and ALROC analyses were performed. SnapPET substantially improved image quality over NAC input for all acquisition times. For 3-s scans, PSNR increased from 28.6 ± 4.3 to 35.3 ± 6.0 and SSIM from 0.85 ± 0.09 to 0.98 ± 0.02, while NMSE decreased from 0.23 ± 0.18 to 0.03 ± 0.01 (all p < 0.001). SUV metrics in lesion ROIs showed high correlation with reference AC PET, with minimal bias. Compared with UNet, GAN, and DDPM, SnapPET achieved the best quantitative metrics and the highest lesion-level ROC and ALROC curves, indicating superior detectability and localization. Reader scores for noise suppression, artifact reduction, contrast retention, and lesion discrimination were consistently higher for SnapPET images, particularly for 3- and 15-s acquisitions. SnapPET enables rapid pediatric PET imaging by generating high-quality, AC-like 2D images directly from ultra-short NAC acquisitions. The method improves image quality, preserves quantitative accuracy, and enhances lesion detectability compared with both NAC input and competing deep-learning baselines. This CT-sparing screening and triage approach has the potential to reduce reliance on attenuation-correction CT and support personalized, lower-dose CT strategies in pediatric lymphoma care.
Journal IF-equivalent: 3.7 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Hu J, Huang Y, Zhang Q, Hu Y, Zhao Y, Zhang X, et al. A rapid total-body PET imaging approach for pediatric patients using non-attenuation-corrected PET scans. EJNMMI Phys. 2026 May 22;13(1):61. doi:10.1186/s40658-026-00889-x. PMID: 42171886; PMCID: PMC13247023.Checked: Full text checked - Abstract
Perforation following chemotherapy in gastrointestinal lymphoma (PFCGL) is a rare but severe and life-threatening complication. Early pre-chemotherapy prediction is crucial for optimizing treatment and improving outcomes, yet it remains challenging. This study aimed to develop a machine learning (ML) model combining clinical features and [ 18 F]FDG PET/CT radiomics signature to predict the risk of PFCGL. A total of 257 patients diagnosed at two hospitals with gastrointestinal lymphoma were included. Clinical data and PET/CT images were collected. PET/CT radiomics signature were extracted from the FDG-avid regions. Five ML algorithms were used to develop predictive models, including clinical, PET/CT radiomics, and clinical-PET/CT models. The SHapley Additive exPlanations (SHAP) method was applied for model interpretability. The internal training, internal validation and external validation cohorts consisted of 144 patients (mean age: 56.90 ± 14.69 years; 88 males), 63 patients (mean age: 61.76 ± 13.71 years; 36 males) and 50 patients (mean age: 63.28 ± 12.17 years; 27 males), respectively. Among these, 22, 11, and 11 patients were diagnosed with PFCGL in the three cohorts, correspondingly. For clinical features, elevated C-reactive protein levels and pathological type of T-cell non-Hodgkin's lymphoma were risk factors for predicting PFCGL. SHAP analysis revealed that PET/CT radiomics signature significantly contributed to the prediction of PFCGL. The clinical-PET/CT model utilizing the logistic regression (LR) algorithm exhibited superior predictive performance for PFCGL compared to the individual clinical model and the PET/CT radiomics model employing the ExtraTrees algorithm. This was evidenced by the area under the curve (AUC) values of 0.918, 0.872 and 0.857 in the training cohort, 0.858, 0.795 and 0.849 in the internal validation cohort, and 0.852, 0.770 and 0.807 in the external validation cohort, respectively. The combined model that integrates [ 18 F]FDG PET/CT radiomics signature with clinical features through the application of machine learning provides a clinically actionable nomogram for the evaluation and prediction the risk of PFCGL.
Journal IF-equivalent: 5.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Chen D, Li T, Cui J, Yang C, Liu Z, Huang H, et al. An interpretable machine learning model integrating [18F]FDG PET/CT radiomics and clinical features for predicting perforation following chemotherapy in gastrointestinal lymphoma: a multicenter study. Eur J Nucl Med Mol Imaging. 2026 May 19;53(10):5745-59. doi:10.1007/s00259-026-07921-3. PMID: 42151619.Checked: Abstract only - Abstract
Background : Diffuse large B-cell lymphoma (DLBCL) is one of the most frequent lymphomas. To date, it is not possible to identify which DLBCL patients will have an aggressive clinical evolution only by using hematoxylin and eosin (H&E) histological images. Methods : This study predicted the prognosis of DLBCL using H&E images, computer vision and deep learning. The series included 114 DLBCL cases, split into 2 prognostic groups according to overall survival, and 44 cases of reactive lymphoid tissue. Results : The curve fitting and slope analysis showed a point of inflection at 2 years (24 months), which differentiated patients with aggressive clinical evolution ("Dead < 2 years", b1 = -0.024) from the rest with moderate clinical evolution ("Others", b1 = -0.003). Twenty different convolutional neural networks (CNNs) were used, and explainable artificial intelligence (XAI) was also applied. The final model based on DarkNet-19 predicted prognosis groups with high performance (test set accuracy = 96.3%). The other performance parameters were precision (94.5%), recall (95.0%), false positive rate (3.1%), specificity (96.9%), and F1 score (94.7%). XAI, including grad-CAM, occlusion sensitivity, and image-LIME, confirmed that the CNN focused on the correct areas. Hybrid partitioning to prevent information leakage with patient-based analysis, image classification between DLBCL and 44 cases of reactive lymphoid tissue, and hyperparameter tuning were also successfully performed. Correlation with the clinicopathological characteristics found that the Dead < 2 years group was correlated with stage III-IV, International Prognostic Index (IPI) High + High/intermediate, progressive disease, non-GCB cell-of-origin, CD10-, BCL2+, and Epstein-Barr virus (EBER)+. Analysis of the microenvironment, immune checkpoint, cell cycle, and germinal center markers showed that Dead < 2 years had higher IL10, PD-L1, and CD163 levels and lower E2F1 protein expression. No differences were found for Ki67, CSF1R, CASP8, TNFAIP8, LMO2, MYC, MDM2, CDK6, and TP53 markers at a quantitative level. Conclusions : The DLBCL overall survival can be predicted using H&E histological images and deep learning using the 2-year (24 months) point (similar to POD24). This trained CNN can be used as a pretrained model for transfer learning in the future.
Journal IF-equivalent: 5.4 (OpenAlex 2-year mean citedness, value as of 2026-10-23, retrieved 2026-10-24; not the official Clarivate IF)Reference: Carreras J. Clinicopathological Characteristics and Prediction of Overall Survival and Death Within 2 Years in Diffuse Large B-Cell Lymphoma Based on Histological Images and Deep Learning. Biomedicines. 2026 May 17;14(5):1134. doi:10.3390/biomedicines14051134. PMID: 42193459; PMCID: PMC13205024.Checked: Full text checked - Abstract
To investigate the feasibility of non-invasively identifying bone marrow involvement (BMI) in follicular lymphoma (FL) using baseline 18 F-FDG PET/CT combined with multidimensional feature fusion, and to compare the impact of different bone marrow volume-of-interest (VOI) frameworks on model performance. This retrospective study included 187 patients with newly diagnosed FL, 93 of whom had BMI. Based on baseline 18 F-FDG PET/CT, two bone marrow VOI frameworks were constructed: a pelvic VOI framework and a spine-pelvis VOI framework. Clinical features, conventional imaging features, radiomic features, and deep learning features were extracted. A hierarchical feature screening strategy was employed: clinical and conventional imaging features were screened using univariate logistic regression, Spearman's correlation analysis, and multivariate logistic regression, whereas high-dimensional radiomic and deep learning features were screened using LASSO regression combined with the Boruta algorithm. Based on the selected features, six different modelling schemes were developed. The optimal scheme was selected using the area under the receiver operating characteristic curve (AUC) in the independent validation set as the primary metric. Under the optimal scheme, the performance of seven machine learning models-logistic regression (LR), support vector machine (SVM), gradient boosting machine (GBM), neural network (NN), random forest (RF), k-nearest neighbours (KNN), and adaptive boosting (AdaBoost)-was further compared. SHAP analysis was used to interpret the key features of the final model and the direction of their contributions. Compared with the non-BMI group, the BMI group was more likely to present with widespread regional lymph node involvement, B symptoms, larger lymph node lesions, as well as lower Hb, higher LDH, lower Apo A, lower eGFR, and higher β2-MG levels (all P < 0.05). Under both VOI frameworks, the BMI group exhibited higher bone marrow FDG uptake intensity and metabolic burden, as reflected by higher values of conventional PET/CT features, including SUVmean, Standard Deviation (PET), RMS, 25th Percentile Value, Median, 75th Percentile Value, TLG, Glycolysis Q2-Q4, SAM, and SUVpeak (all P < 0.05). Multivariate logistic regression analysis indicated that regional lymph node involvement and β2-MG consistently remained independent predictors across both VOI frameworks, whereas SUVmean retained statistical significance only within the pelvic VOI framework. A comparison of six modelling schemes revealed that the scheme integrating the spine-pelvis VOI framework with clinical features, conventional imaging features, and radiomic features performed best. Under this scheme, the GBM model achieved the best overall performance on the independent validation set (AUC = 0.906, Accuracy = 0.877, Precision = 0.926, Sensitivity = 0.833, Specificity = 0.926, F1 score = 0.877). SHAP analysis revealed that, in addition to LNr (≥ 5) and β2-MG, first-order statistical features such as PET-Orig-FO-IQR, as well as texture features derived from wavelet/LBP transformations-including PET-Wav-HLL-NGTDM-Strength, PET-Wav-HLL-GLRLM-SRHGLE, CT-LBP3D-m1-GLCM-MCC, and PET-LBP3D-m2-GLSZM-SAHGLE-also made significant contributions. These findings suggest that BMI-associated imaging phenotypes are characterised not only by increased bone marrow metabolism but also by remodelling of the grey-level distribution and spatial heterogeneity within the bone marrow. Bone marrow involvement in follicular lymphoma is associated with higher tumour burden and altered metabolic heterogeneity within the bone marrow. A PET/CT-based radiomic-clinical model showed good performance for non-invasive BMI prediction, and the spine-pelvis VOI framework outperformed the pelvic VOI framework alone. The final GBM model may provide a feasible imaging biomarker for complementary baseline assessment of BMI in FL.
Journal IF-equivalent: 5.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Li SC, Xu YP, He F, Chen QL, Li AM, Fan X, et al. An interpretable PET/CT-based radiomic-clinical model for predicting bone marrow involvement in follicular lymphoma: comparison of pelvic and spine-pelvis VOI frameworks. Eur J Nucl Med Mol Imaging. 2026 May 16;53(9):5475-90. doi:10.1007/s00259-026-07918-y. PMID: 42142271; PMCID: PMC13315510.Checked: Full text checked - Abstract
Objectives : Marginal zone lymphoma (MZL) is a heterogeneous indolent B-cell lymphoma, and current clinical prognostic systems remain limited in identifying transformation risk and refining risk stratification. This study aimed to evaluate the prognostic relevance of artificial intelligence (AI)-quantified immunohistochemical (IHC) markers in MZL and to explore a revised MZL-IPI model. Methods : We retrospectively analyzed 146 patients with pathologically confirmed MZL treated at the Second Xiangya Hospital of Central South University from January 2015 to June 2022. Among them, 111 patients had digitized IHC slides available for AI-assisted quantitative analysis. AI-quantified IHC marker expression was assessed in relation to clinical features, histologic transformation, and survival outcomes. Prognosis-related markers were dichotomized using optimal cut-off values. Survival differences were evaluated using the log-rank test, independent prognostic factors were identified by multivariable Cox regression, and model performance was assessed by cross-validation. Results : Age, B symptoms, hypertension, diabetes mellitus, and gastric involvement were associated with selected IHC parameters. CD3 was independently associated with histologic transformation, with expression below 25.60% indicating higher transformation risk. High CD21 expression independently predicted favorable overall survival (OS), whereas high CD3 expression was associated with inferior progression-free survival (PFS). Incorporating CD21 into the MZL International Prognostic Index (MZL-IPI) improved OS prediction in this cohort. Conclusions : AI-assisted quantitative IHC analysis may provide complementary prognostic information in MZL. The CD21-revised MZL-IPI represents an exploratory framework for integrating AI-derived tissue biomarkers with clinical risk stratification, but external multicenter validation is required before clinical application.
Journal IF-equivalent: 4.0 (OpenAlex 2-year mean citedness, value as of 2026-10-29, retrieved 2026-10-31; not the official Clarivate IF)Reference: Zhang Q, Deng Z, Yan W, Zhu H, Tian Z, Jiang Y, et al. AI-Enhanced Quantitative IHC Analysis for Prognostic Stratification in Marginal Zone Lymphoma: Development of a Revised MZL-IPI Model. Diagnostics (Basel). 2026 May 11;16(10):1456. doi:10.3390/diagnostics16101456. PMID: 42196822; PMCID: PMC13205237.Checked: Full text checked - Abstract
Primary gastric diffuse large B-cell lymphoma (PG-DLBCL) exhibits heterogeneous outcomes, and conventional prognostic systems often fail to capture its unique clinicopathological features. We aimed to develop a robust, interpretable prognostic model to improve risk stratification and guide individualized management. Data from 3773 PG-DLBCL patients (2000-2021) were extracted from the SEER database. Four complementary feature selection strategies-LASSO, Boruta, backward stepwise elimination, and best subset regression (BSR)-were employed to identify stable prognostic variables. Four machine learning (ML) algorithms (logistic regression, support vector machine, k-nearest neighbor, and XGBoost) were trained using these selected variables. Model performance was evaluated through discrimination (AUC), calibration, decision curve analysis, and internal validation. Shapley Additive Explanations (SHAP) were applied for interpretability, and patients were stratified into high- and low-risk groups. Age, stage, chemotherapy, marital status, and income emerged as key prognostic determinants. The XGBoost model based on BSR-selected predictors achieved the highest performance with good calibration and net clinical benefit. SHAP analysis demonstrated that older age, advanced stage and absence of chemotherapy increased predicted risk, whereas marital status and higher income were protective. Risk stratification effectively distinguished survival outcomes in training and testing cohorts (p < 0.001). A web-based tool was developed for individualized risk assessment. We established an interpretable, high-performing ML-based prognostic model for PG-DLBCL that integrates clinical, treatment, and socioeconomic factors. This tool enables precise risk stratification, supports individualized therapeutic decisions, and provides a methodological framework for prognostic modeling in this rare extranodal lymphomas.
Journal IF-equivalent: 5.1 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Lin J, Wang H, Lin H, Xu C. Multi-strategy feature selection and multi-model machine learning for prognostic prediction in primary gastric diffuse large B-cell lymphoma. Digit Health. 2026 May 7;12:20552076261450466. doi:10.1177/20552076261450466. PMID: 42136918; PMCID: PMC13167268.Checked: Full text checked - Abstract
This study explored a machine learning approach using 18 F-FDG PET/CT as a non-invasive alternative to biopsy, incorporating tumor-to-liver ratio (TLR) PET radiomics, and performed survival analysis to improve lymphoma management. In this cohort study, baseline 18 F-FDG PET/CT scans of newly diagnosed, histologically confirmed lymphoma patients were analyzed. Lesions were segmented using 3D Slicer, and radiomic features were extracted and normalized by tumor-to-liver ratios. Patient-level features were used to train three machine learning models (XGBoost, AdaBoost, Logistic Regression) using nested cross-validation with SMOTE for class balancing. A model based on SUV max metrics served as baseline. Radiomic features were also evaluated for correlation with 3- and 5-year survival using the Mann-Whitney U test. A total of 156 lymphoma patients were analyzed, with 2,076 lesions segmented and 200 radiomic features extracted. For subtype classification, AdaBoost achieved the highest AUC for Diffuse Large B-cell (DLBCL) (0.863, accuracy 0.742), while XGBoost performed best for High-Grade Non-Hodgkin lymphoma (NHL) (AUC 0.825, accuracy 0.735) and Nodular Sclerosis Hodgkin Lymphoma (NS-HL) (AUC 0.827, accuracy 0.832). Logistic regression showed the best results for Classical Hodgkin Lymphoma (C-HL) (AUC 0.849, accuracy 0.775). The SUV max -based model (LR-SUV_MAX) consistently underperformed (AUCs: C-HL 0.630, High-Grade NHL 0.700, NS-HL 0.638, DLBCL 0.664), with all differences being statistically significant ( p < 0.001). Radiomic and clinical features including SUV-GLSZM small area emphasis ( p = 0.0019), age ( p = 0.0002), and spleen involvement ( p = 0.0014) were significantly associated with 3- and 5-year overall survival in 110 and 74 patients, respectively. Radiomic features combined with machine learning significantly improve lymphoma subtype classification over SUV max alone and show potential for predicting patient survival. The online version contains supplementary material available at https://doi.org/10.1007/s13139-026-01017-4.
Journal IF-equivalent: 2.0 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Hasanabadi S, Aghamiri SMR, Abin AA, Vosoughi H, Emami F, Bakhshayesh Karam M, et al. Enhanced Lymphoma Subtype Classification and Prognosis Using Machine Learning with 18F-FDG PET/CT Radiomics: Beyond SUVmax. Nucl Med Mol Imaging. 2026 Apr 27;60(5):566-80. doi:10.1007/s13139-026-01017-4. PMID: 42824644.Checked: Abstract only - Abstract
PURPOSE: The [18F]FDG-PET-derived total metabolic tumor volume (TMTV) has a high prognostic value in patients with Hodgkin and Non-Hodgkin lymphoma. However, in order to enable TMTV as a biomarker for clinical use, an accurate and fast method of tumor delineation in lymphoma patients is needed. Deep-learning-based methods have shown promising results in this field and offer distinct advantages over classical approaches. Therefore, the goal of this work was to train a convolutional neural network (CNN) for delineation of all lymphoma lesions regardless of their size and uptake characteristics while performing the optimal contouring of each individual lesion. METHODS: A neural network was trained with the nnU-Net software package. A total of 1192 [18F]FDG-PET/CT scans from 716 patients with Non-Hodgkin lymphoma participating in the PETAL trial comprised the main dataset which was used for training. The ground truth delineation included all lesions that were clinically considered as lymphoma manifestations by an experienced observer and was developed iteratively with the assistance of intermediate CNN models. Performance of the trained network was assessed in the main dataset via 5-fold cross-validation as well as in external benchmark dataset (N = 60 scans). RESULTS: Comparing the manual and automated delineations in the main (external) dataset, the aggregated Dice coefficient reached 0.895 (0.715) and the corresponding TMTVs were highly correlated with R2 = 0.974 (0.767). The main (external) dataset contained a total of 8971 (713) manually delineated lesions, the detection sensitivity of which was 71.2% (77.6%) with the positive predictive value of 84.1% (63.1%). Univariate Cox regression analysis in the main dataset revealed both manually and automatically derived TMTVs as highly prognostic factors for progression-free survival with very similar hazard ratios (HR = 3.5; p < 0.001 and HR = 3.7; p < 0.001, respectively). CONCLUSION: In this study we presented a CNN model capable of accurate TMTV delineation in [18F]FDG-PET/CT images of lymphoma patients. It is trained to delineate all tumor lesions while accounting for typical caveats inherent in this task. The developed neural network allows for substantial acceleration of quantitative analysis of lymphoma imaging data and has the potential for supervised clinical use.
Journal IF-equivalent: 5.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Nikulin P, Hoberück S, Apostolova I, Maus J, Hüttmann A, Dührsen U, et al. A convolutional neural network for fully automated total metabolic tumor volume delineation in patients with aggressive Non-Hodgkin lymphoma. Eur J Nucl Med Mol Imaging. 2026 Mar 24;53(8):5074-87. doi:10.1007/s00259-026-07810-9. PMID: 41872336; PMCID: PMC13249771.Checked: Full text checked - Abstract
Lymphoma-associated haemophagocytic lymphohistiocytosis (LA-HLH) is associated with a high mortality rate, making early diagnosis and appropriate treatment critical for improving patient outcomes. In this study, we enrolled 126 patients diagnosed with LA-HLH and 254 with non-LA-HLH. A machine learning-based predictive model was developed and validated to enable timely differentiation of LA-HLH from other HLH subtypes. The model incorporated 11 predictive variables for early LA-HLH prediction, among which the top five most influential were age, ferritin, monocyte percentage, haemoglobin and platelet count. Among seven machine learning algorithms evaluated, the random forest model demonstrated the best performance, achieving an area under the curve (AUC) of 0.946 on the training set and 0.794 on the validation set. We subsequently evaluated combined models that incorporated disease-specific indicators such as soluble interleukin-2 receptor (sCD25) and PET-CT SUVmax, which resulted in a non-significant increase in AUC on the validation set. Finally, the optimal model was deployed as a web-based tool to support early aetiological differentiation. This may facilitate prompt initiation of targeted examination and appropriate treatment for patients with LA-HLH.
Journal IF-equivalent: 3.0 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Luan M, Jia R, Wang D, Zhang F, Han X, Sheng X, et al. Early prediction of adult lymphoma‐associated haemophagocytic lymphohistiocytosis using an interpretable machine learning model. Br J Haematol. 2026 Mar 23;208(6):2047-58. doi:10.1111/bjh.70447. PMID: 41871894.Checked: Abstract only - Abstract
OBJECTIVE: To develop and validate a prognostic imaging biomarker derived from baseline [¹⁸F]FDG PET/CT using tabular deep learning for prediction of progression of disease within 24 months (POD24) and survival risk stratification in patients with follicular lymphoma (FL). METHODS: This retrospective multicenter study included 309 patients with newly diagnosed FL (grades 1-3a) from five independent medical centers. Tumor volumes segmented from baseline [¹⁸F]FDG PET and CT images were used to extract high-throughput radiomic features. Five conventional machine learning algorithms and four advanced tabular deep learning models were developed and compared. The predictive output of the GAMformer model was defined as the deep learning score (DLS). The DLS was integrated with clinical variables and PET metabolic parameters to construct a multiparametric model in the training cohort. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis, and further validated in validation cohort. RESULTS: During a median follow-up of 44 months, POD24 occurred in 55 patients. The DLS demonstrated strong predictive performance for POD24 (training AUC = 0.857; validation AUC = 0.753). The multiparametric model further improved discrimination, achieving AUCs of 0.882 in the training cohort and 0.797 in the validation cohort, outperforming FLIPI, FLIPI-2, and PRIMA-PI. Calibration showed good agreement, and decision curve analysis indicated higher net clinical benefit. The DLS stratified survival risk (P < 0.05) and remained predictive of survival and POD24 across histologic grades. CONCLUSIONS: The DLS derived from baseline [¹⁸F]FDG PET/CT enables POD24 prediction and accurate survival risk stratification in FL, supporting its potential role in precision management.
Journal IF-equivalent: 5.5 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Huang H, Zhang T, Jiang Z, Zhang Z, Jiang Q, Teng Y, et al. A deep learning PET/CT biomarker for early progression (POD24) and survival stratification in follicular lymphoma: a multicenter study. Eur J Nucl Med Mol Imaging. 2026 Mar 23;53(8):5060-73. doi:10.1007/s00259-026-07838-x. PMID: 41870550.Checked: Abstract only - Abstract
Histone deacetylase 2 (HDAC2) plays a critical role in the pathogenesis of diffuse large B-cell lymphoma (DLBCL), positioning it as an attractive therapeutic target. In this study, we applied an integrated computational strategy that combined machine learning-based quantitative structure-activity relationship (QSAR) modelling, molecular docking, ADMET filtering, and molecular dynamics (MD) simulations to identify and optimize potential HDAC2 inhibitors. A dataset of 1995 HDAC2-active molecules was assembled and reduced to 25 key molecular descriptors, enabling the development of a robust Random Forest QSAR model (R 2 = 0.926, CCC = 0.956). Top-predicted compounds underwent docking and pharmacokinetic evaluation, highlighting compound 10 as a promising lead (binding energy ≈ -10.2 kcal·mol -1 ). Guided optimization produced analogue 10 g, which displayed enhanced affinity (-10.9 kcal·mol -1 ), stable protein-ligand interactions in MD simulations, and favourable MM-GBSA binding free energy (ΔG_bind ≈ -40.3 kcal·mol -1 ). In silico pharmacokinetic analysis further suggested acceptable safety and drug-likeness. This multi-stage pipeline prioritizes 10 g as a strong candidate for experimental validation and demonstrates the value of integrating predictive modelling with structure-based refinement in anti-lymphoma drug discovery.
Journal IF-equivalent: 3.1 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Isah JJ, Uzairu A, Uba S, Ibrahim MT. Integrating machine learning and molecular simulations for the design of potent HDAC2 inhibitors in diffuse large B-cell lymphoma. Comput Biol Chem. 2026 Aug;123:108949. doi:10.1016/j.compbiolchem.2026.108949. PMID: 41722288.Checked: Abstract only - Abstract
Clinical target volume (CTV) delineation for involved-site radiation therapy (ISRT) in Hodgkin lymphoma (HL) is time-consuming because of the need to analyze multi-timepoint positron emission tomography (PET)/computed tomography (CT) scans coregistered to the planning CT. Our goal was to develop automated CTV segmentation algorithms that integrated multimodality imaging to facilitate ISRT planning. This study included planning CT, baseline PET/CT (PET1), and interim PET/CT (PET2) scans from 288 pediatric patients with high-risk HL enrolled in the Children's Oncology Group AHOD 1331 trial. Data from 58 patients across 24 institutions were held out for external testing, while the remaining 230 cases from 95 institutions were used for model development. We investigated 3 deep learning (DL) architectures (SegResNet, ResUNet, and SwinUNETR) and evaluated the impact of incorporating PET1 and PET2 images along with the planning CT. Performance was assessed using the Dice similarity coefficient (DSC) and 95% Hausdorff distance (HD95). Interobserver variability was estimated by comparing original institutional CTVs with those newly delineated by 4 radiation oncologists on 10 cases. The quality of CTVs generated by the top-performing model was independently assessed by radiation oncologists in 40 other cases using a 5-point Likert scale and compared against the original institutional CTVs. In the external cohort, a SwinUNETR model incorporating planning CT, PET1, and PET2 images achieved the highest performance, with a DSC of 0.72 and HD95 of 34.43 mm. All models incorporating PET/CT images were significantly better (P < .01) than planning CT-only models. Interobserver variability analysis yielded a DSC of 0.70 and HD95 of 30.14 mm. In clinical evaluation, DL-generated CTVs received a mean quality score of 3.38 out of 5, comparable to original physician-delineated CTVs (3.13; P = .13) CONCLUSIONS: The DL model was able to generate clinically useful CTVs with quality comparable to manually delineated CTVs, suggesting its potential to improve physician efficiency in ISRT planning.
Journal IF-equivalent: 0.9 (OpenAlex 2-year mean citedness, value as of 2026-11-04, retrieved 2026-11-05; not the official Clarivate IF)Reference: Tie X, Milgrom SA, Lo AC, Charpentier AM, LaRiviere MJ, Maqbool D, et al. Multimodality Artificial Intelligence for Involved-Site Radiation Therapy: Clinical Target Volume Delineation in High-Risk Pediatric Hodgkin Lymphoma. Int J Radiat Oncol Biol Phys. 2026 Jul;125(3):919-30. doi:10.1016/j.ijrobp.2025.12.005. PMID: 41391612; PMCID: PMC12991415.Checked: Abstract only