AI and Cancer Research ── 2026-10-11
New papers on cancer and AI, collected from journal feeds, PubMed (searched across all journals) and OpenAlex (a public index used instead of Google Scholar), judged by Jev (an AI that makes language judgements): original research, AI central, which of 13 themes, treatment or trials, and then ranked by importance up to a daily limit. Journals are not filtered by impact factor; each paper shows its journal's IF-equivalent (OpenAlex two-year mean citedness) and the date of that value. Each entry shows the full original abstract.
Read by cancer type: Cancer biology（2） · Breast（2） · Lung（1） · Pancreatic（0） · Prostate（2） · ACC（0） · Brain（1） · Ovarian（0） · Endometrial（0） · Gastric（0） · Liver（2） · Kidney（0） · Bladder（0）
Cancer molecular biology and AI
- Abstract
Background: Lung cancer is one of the most lethal malignancies globally. However, accurately differentiating subpleural malignant lesions from benign infectious conditions via grayscale ultrasound remains difficult due to operator subjectivity. We developed and validated a two-stage deep learning framework for automated lesion segmentation and benign-malignant classification. Materials and methods In this retrospective study, the research data were derived from patients with SPLs who attended Center 1 between Jan 2023–Dec 2024, as well as patients recruited from Center 2 and Center 3 Jun 2024–Dec 2024 (serving as External Test Set 1 and 2, respectively). The nnU-Net, U-Net and DeepLabv3 + models were adopted for lesion segmentation, while the DenseNet121/201, EfficientNet-B0/B1/B2/B3/B4/B5, ResNet18/50/101, Inception-V3 and VGG19 models were used for benign-malignant classification. Model performance was evaluated using metrics including the Dice similarity coefficient and area under the receiver operating characteristic curve (AUC). Diagnostic performance of 6 radiologists (3 junior, 3 senior) with varying experience was compared between unaided and AI-assisted readings; pooled three-reader differences were estimated by case-cluster bootstrap with Bonferroni adjustment.
Results: A total of 1059 patients were included in the study (61.32 ± 13.82 years; 726 male patients). The nnU-Net model achieved optimal segmentation (Dice: 0.920 [tuning-validation set, n = 257], 0.913 [External Test Set 1, n = 108], 0.915 [External Test Set 2, n = 94]). Among thirteen candidate classification architectures, DenseNet121 yielded the best performance on the internal tuning-validation set and was therefore selected as the final model; it was then evaluated on two external cohorts, achieving AUCs of 0.889 (95% CI: 0.827–0.951) and 0.865 (95% CI: 0.790–0.940) in External Test Sets 1 and 2, respectively. Overall diagnostic accuracy was significantly higher in the AI-assisted reading session for junior radiologists in both external test cohorts (both Bonferroni- adjusted P < 0.001), whereas improvements among senior radiologists did not reach statistical significance after multiplicity adjustment.
Conclusion: The deep learning model constructed in this study can accurately realize automatic segmentation and benign-malignant classification of SPL ultrasound images, serving as a potential auxiliary diagnostic tool. Trial registration Not applicable.
Journal IF-equivalent: 3.8 (OpenAlex 2-year mean citedness, value as of 2026-10-02, retrieved 2026-10-03; not the official Clarivate IF)Reference: Ma Q, Liang T, Yi J, Bai H, Li Y, Shen M, et al. A grayscale ultrasound-based two-stage deep learning framework for automatic segmentation and benign-malignant differentiation of subpleural pulmonary lesions. BMC Med Imaging. 2026 Oct 9 [Epub ahead of print]. doi:10.1186/s12880-026-02887-y.Checked: Abstract only - Abstract
Predicting how cancers respond to treatment remains difficult because tumors are diverse and computational models often perform poorly on unfamiliar drugs or samples. Here we show that the accuracy of deep learning models varies across molecular inputs, drug descriptions, model designs, and evaluation settings, with the largest losses when models are tested on previously unseen drugs or distinct datasets. We therefore develop Drug Response Integration and Voting Ensemble, a meta-learning framework that combines models using predictions from held-out data. A nine-model ensemble improves predictive accuracy, drug ranking, and stability over individual models and ensemble approaches. We implement the framework as a reproducible Nextflow workflow and use it to build a resource of measured and predicted drug responses across cancer cell lines, patient-derived samples, and compound libraries. In colorectal cancer, the framework identifies treatments with confirmed antitumor activity in cell and animal models. This approach provides a scalable strategy for therapeutic discovery. We benchmark drug response models and develop DRIVE, a meta-learning ensemble for large-scale cancer drug sensitivity prediction. DRIVE improves robustness and prioritizes compounds with validated antitumor activity in colorectal cancer.
Journal IF-equivalent: 6.6 (OpenAlex 2-year mean citedness, value as of 2026-10-10, retrieved 2026-10-11; not the official Clarivate IF)Reference: Feng Y, Zhou S, Luo S, Feng B, Wu H, Zeng Y, et al. An ensemble framework for robust drug response prediction and large-scale sensitivity profiling across cancers. Commun Biol. 2026 Oct 8 [Epub ahead of print]. doi:10.1038/s42003-026-11124-9.Checked: Abstract only
See all 2 Cancer biology papers →
Breast cancer: treatment, trials and AI
- Abstract
Preoperative evaluation of sentinel lymph node (SLN) metastasis facilitates individualized axillary management for patients with invasive breast cancer (IBC). This study aimed to develop and externally validate a deep learning-derived ultrasound radiomics model to predict SLN metastasis. In this retrospective two-center study, 246 pathologically confirmed IBC patients were enrolled. 194 patients from Center A were split into training ( n = 155) and internal test ( n = 39) cohorts, while 52 patients from Center B served as the external validation cohort. Deep features were extracted from preoperative ultrasound images via ResNet50. After feature selection within the training cohort, a gradient boosting decision tree classifier was constructed. Model discrimination was assessed using receiver-operating characteristic analysis, and decision-curve analysis was applied for exploratory net-benefit evaluation. Twenty-two deep learning-derived features were retained. The model achieved areas under the receiver operating characteristic curve (AUC) of 0.734 (95% CI, 0.655–0.813), 0.761 (95% CI, 0.601–0.920), and 0.742 (95% CI, 0.598–0.886) in the training, internal test, and external validation cohorts, respectively. In the external cohort, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were 0.769, 0.650, 0.844, 0.722, and 0.794, respectively. Decision curves were exploratory and do not establish clinical utility. The model showed moderate discrimination for SLN metastasis, with similar AUC point estimates but wide confidence intervals across cohorts; it provides preliminary discrimination and risk ranking only and is not currently clinically actionable. Prospective validation in larger cohorts is required before any clinical use.
Journal IF-equivalent: 3.8 (OpenAlex 2-year mean citedness, value as of 2026-10-02, retrieved 2026-10-03; not the official Clarivate IF)Reference: He K, Qiu Y, Zeng B, Huang Z, Lin J, Chen J, et al. Ultrasound-based deep learning-derived radiomics for preoperative prediction of sentinel lymph node metastasis in invasive breast cancer: a two-center retrospective study. BMC Med Imaging. 2026 Oct 9 [Epub ahead of print]. doi:10.1186/s12880-026-02919-7.Checked: Abstract only - Abstract
Accurate assessment of HER2 status in breast cancer has been critical for guiding therapy and has become even more important with the emergence of antibody–drug conjugates, now also indicated in HER2-low tumors. However, inter- and intraobserver variability limits the reproducibility of HER2 IHC scoring among pathologists. Artificial intelligence (AI) models offer potential to standardize and improve diagnostic accuracy and bring new insights into current practices shortcomings. We conducted a study recruiting generalist and specialist pathologists from Rede D’Or centers across Brazil to assess digitized HER2 IHC whole slide images. The same images were presented for the pathologists with an interval of one month and to the AIM-HER2 (PathAI ®, Boston, MA) AI model. Intra- and interobserver agreement, as well as agreement with AI, were measured across 126 breast cancer samples. The association between sample features and agreement metrics was also analyzed using AI spatial breakdown data. Among pathologists, the median intraobserver agreement was 66.67%, and the median agreement with AI was 60.8%. Median interobserver agreement was 67.65%, with high agreement (> 85%) in 25.4% of samples. Significant positive correlations were observed among all agreement metrics. Samples with lower intra-sample heterogeneity, as determined by AI spatial breakdown scores, were associated with higher agreement levels. Our findings highlight significant variability in HER2 IHC scoring among pathologists. We found that AI-assessed intra-sample heterogeneity is correlated with a lower agreement rate among pathologists. We believe this shows a potential limitation of current scoring practices and that AI models such as AIM-HER2 can have a role in increasing reproducibility of analysis by indicating the samples that are more likely to result in diagnostic discordance between evaluators.
Journal IF-equivalent: 3.9 (OpenAlex 2-year mean citedness, value as of 2026-10-09, retrieved 2026-10-10; not the official Clarivate IF)Reference: Ferrari P, Petaccia de Macedo M, Werneck da Cunha I, Soares-Souza GB, Soares FA, Breast Cancer Cooperative Study Group, et al. Artificial Intelligence Model’s assessment of intra-sample heterogeneity of HER2 IHC in breast cancer is related to interobserver and intraobserver agreement among pathologists. BMC Cancer. 2026 Oct 9 [Epub ahead of print]. doi:10.1186/s12885-026-17049-0.Checked: Abstract only
Lung cancer: treatment, trials and AI
- Abstract
Limiting acute esophagitis remains a clinical challenge in the treatment of locally advanced non-small cell lung cancer using chemoradiotherapy. In this study, machine learning (ML) algorithms were used to predict acute esophagitis in patients (n = 451; training set 70% and independent test set 30%) treated in the Radiation Therapy Oncology Group 0617 clinical trial. Multiple ML models were trained to predict grade ≥ 2 esophagitis using clinical, radiomics, and dosimetric/dosiomics features. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). The best-performing model was explained using the Shapley additive explanation framework and decision curve analysis. The highest-accuracy model was based on radiomics–dosiomics features using the least absolute shrinkage and selection operator model (AUC = 0.70; 95% confidence interval (CI): 0.60–0.78). A novel model that accounts for esophageal geometry relative to tumor location achieved an AUC of 0.67 (95% CI: 0.58–0.76) using dose–volume histogram relationships. The ML models revealed important relationships that can guide radiation dose planning to minimize individual risk and facilitate personalized radiation therapy.
Journal IF-equivalent: 1.5 (OpenAlex 2-year mean citedness, value as of 2026-10-10, retrieved 2026-10-11; not the official Clarivate IF)Reference: Upadhaya T, Chetty IJ, Lui J, Atkin KM. Characterization of radiation-induced esophagitis using explainable machine learning algorithms for lung cancer patients treated in the NRG/RTOG 0617clinical trial. AIH. 2026 Oct 8;0(0):026220054. doi:10.36922/aih026220054.Checked: Abstract only
Pancreatic cancer: treatment, trials and AI
No new qualifying paper for this issue.
Prostate cancer: treatment, trials and AI
- Abstract
Prostate cancer (PCa) exhibits profound clinical heterogeneity, and current prognostic tools fail to capture the molecular drivers of aggressive progression. The functional interplay between E2F-driven proliferation and G2/M checkpoint regulation forms a core axis connecting cell cycle dysregulation with dependence on DNA damage repair, yet integrated signatures capturing this crosstalk remain lacking. We developed a customized AutoML R package to systematically construct and benchmark prognostic models across multiple independent PCa cohorts. An integrated E2F-G2M signature was established and validated across five independent cohorts comprising 1,128 patients. Genomic characterization, drug sensitivity prediction, and experimental validation were performed to investigate the biological features and therapeutic implications of the E2F-G2M signature. Both E2F and G2/M pathways were consistently activated during PCa progression and independently predicted poor outcomes. The integrated E2F-G2M signature achieved a mean C‑index of 0.78 across four external PCa cohorts, outperforming single‑pathway models, conventional clinical variables, and previously reported signatures. E2F-G2M–high tumors were associated with recurrent tumor suppressor gene alterations, attenuated AR signaling, and neuroendocrine-like features. Additionally, E2F-G2M-high tumors showed sensitivity to PARP inhibitors and selected replication stress-targeting agents, including ATR, WEE1, and CHK1/2 inhibitors, while exhibiting reduced sensitivity to CDK4/6 and ATM inhibitors. Combination strategies targeting E2F-associated cell-cycle programs and G2/M checkpoint regulation exhibited context-dependent synergy related to RB1 status. The integrated E2F-G2M signature provides a robust prognostic framework that captures aggressive PCa phenotypes and identifies potential therapeutic vulnerabilities.
Journal IF-equivalent: 10.3 (OpenAlex 2-year mean citedness, value as of 2026-10-10, retrieved 2026-10-11; not the official Clarivate IF)Reference: Wang L, Wang L. A machine learning–derived E2F-G2M signature for prognostic stratification and therapeutic targeting in aggressive prostate cancer. Cell Commun Signal. 2026 Oct 9 [Epub ahead of print]. doi:10.1186/s12964-026-03276-2.Checked: Abstract only - Abstract
Objective: Biochemical recurrence (BCR) after laparoscopic radical prostatectomy (LRP) significantly affects the long-term survival of prostate cancer (PCa) patients. This study aimed to develop and validate an interpretable machine learning model integrating clinicopathological characteristics and preoperative inflammatory and nutritional indices to predict recurrence-free survival (RFS) and explore potential molecular therapeutic targets.
Methods: A retrospective cohort of 320 PCa patients undergoing LRP at Beijing Chaoyang Hospital, Capital Medical University was analyzed. We integrated demographic, clinicopathological variables, and seven composite inflammatory/nutritional indices before LRP. A Mime-based machine learning survival pipeline was employed, combining Cox-based feature selection with survival algorithms. Model performance was evaluated using the concordance index (C-index), time-dependent ROC, calibration curves, and decision curve analysis (DCA). SHAP (SHapley Additive exPlanations) was used for model interpretation and deriving clinical cutoffs. Furthermore, intersecting genes of BCR, lymphocyte-to-monocyte ratio (LMR), and prognostic nutritional index (PNI) were identified via GeneCards, followed by drug sensitivity screening and molecular docking.
Results: The StepCox[both] combined with Random Survival Forest model demonstrated optimal performance, achieving a validation C-index of 0.788. SHAP analysis identified the top five predictive factors for recurrence: preoperative LMR, PNI score, postoperative Gleason score ≥ 8, positive surgical margin, and lymphovascular invasion. Model-derived cutoffs were determined as LMR = 3.9 and PNI = 48.8. Intersection analysis revealed 1211 shared targets among BCR, LMR, and PNI, with TP53 and PTEN ranking highest. Drug sensitivity analysis and molecular docking suggested potential inhibitory effects of compounds like sabutoclax and dordaviprone on these targets.
Conclusions: The proposed model may serve as a robust and interpretable risk stratification for postoperative PCa patients. The integration of clinical phenotyping with exploratory molecular docking offers a translational pathway from prognostic indices to potential therapeutic hypotheses, warranting further prospective validation.
Journal IF-equivalent: 3.5 (OpenAlex 2-year mean citedness, value as of 2026-10-10, retrieved 2026-10-11; not the official Clarivate IF)Reference: Wang H, Zhang Y, Li Z, Wu J, Cao H, Gan L, et al. Interpretable Machine Learning Prognostic Model Integrating Inflammatory-Nutritional Biomarkers for Post-Prostatectomy Biochemical Recurrence: Development, Internal Validation, and Translational Target Exploration. Curr Oncol. 2026 Oct 9;33(10):608. doi:10.3390/curroncol33100608.Checked: Abstract only
Adenoid cystic carcinoma: treatment, trials and AI
No new qualifying paper for this issue.
Brain tumours: treatment, trials and AI
- Abstract
Background/Objectives: Adult-type diffuse gliomas (ADGs) are aggressive brain tumors with highly variable clinical outcomes. Prognostic assessment directly from routine hematoxylin and eosin (H&E)-stained whole-slide images (WSIs) could support personalized decision-making.
Methods: To our knowledge, this is the first systematic benchmarking of pathology foundation models (FMs) and multiple instance learning (MIL) methods for overall survival (OS) risk stratification in ADGs under the 2021 World Health Organization (WHO) classification. A retrospective, multi-institutional cohort of 817 subjects from The Cancer Genome Atlas (TCGA) was analyzed. Following tissue segmentation and patch-level curation, we evaluated 77 pathology FM-MIL combinations (11 FMs × 7 MIL methods), together with seven ResNet-50 baseline configurations (84 encoder-MIL configurations in total), using patient-level 10-fold cross-validation. A perturbation-based evaluation assessed attention mechanism faithfulness.
Results: UNI2 paired with MambaMIL achieved the highest cross-validated concordance index (C-index = 0.77), whereas its time-dependent AUC was modest (0.60–0.61 at 12–36 months), and several combinations performed comparably. Kaplan–Meier analysis demonstrated significant separation between high- and low-risk groups (log-rank p = 9.16 × 10−17; hazard ratio [HR] = 2.90, 95% confidence interval [CI]: 2.23–3.76), with a median OS of 16.6 months (95% CI: 15.0–20.4) versus 63.5 months (95% CI: 43.9-NR), respectively. Tertile stratification confirmed a monotonic OS gradient from 94.5 months to 14.7 months (p = 3.25 × 10−17). Risk stratification within individual WHO 2021 types was not significant. Most domain-specific FMs outperformed the ImageNet-pretrained baseline, and the removal of highly attended patches reduced performance, supporting attention faithfulness.
Conclusions: The AI-derived risk score remained associated with OS after adjustment for individual clinical and molecular variables, but independent prognostic value was not demonstrated in the fully adjusted model (HR = 1.32, 95% CI: 0.94–1.85, p = 0.104), and its discrimination largely reflects differences between WHO 2021 types. This benchmarking framework provides guidance for FM-MIL design in computational pathology applications targeting survival prediction; external validation is required before clinical application.
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: Innani S, Pitarch-Abaigar C, Majeed MM, Harmsen H, Bell WR, Makris D, et al. AI-Based Prognostic Risk Stratification of Adult-Type Diffuse Glioma Patients from H&E-Stained Slides. Cancers. 2026 Oct 9;18(20):3259. doi:10.3390/cancers18203259.Checked: Abstract only
Ovarian cancer: treatment, trials and AI
No new qualifying paper for this issue.
Endometrial cancer: treatment, trials and AI
No new qualifying paper for this issue.
Gastric cancer: treatment, trials and AI
No new qualifying paper for this issue.
Liver cancer: treatment, trials and AI
- Abstract
Predicting transarterial chemoembolization (TACE) refractoriness is critical for optimizing treatment strategies in hepatocellular carcinoma (HCC); however, accurate risk assessment remains challenging. The initial therapeutic response, captured through imaging and clinical indicators, reflects intrinsic treatment sensitivity and holds substantial predictive value for long-term refractoriness. This study evaluated treatment effect heterogeneity following the initial TACE to develop a multimodal nomogram for predicting TACE refractoriness after repeated sessions. Using multiparametric MRI spatial habitat radiomics, we identified three imaging-defined habitats with signal profiles resembling liquefactive necrosis, residual viable tumor, and coagulative necrosis. By integrating habitat and conventional radiomic features, an interpretable radiomic score was constructed utilizing eight machine learning algorithms. This score was subsequently combined with longitudinal dynamic clinical parameters to build a joint clinical-radiomic nomogram. The nomogram achieved areas under the curve (AUCs) of 0.924, 0.860, and 0.864 in the training, internal validation, and external validation cohorts, respectively. These findings suggest that integrating habitat radiomics with dynamic clinical features may help stratify the risk of subsequent TACE refractoriness after the initial TACE session and before repeated treatment, potentially supporting early treatment-response assessment and individualized clinical decision-making.
Journal IF-equivalent: 5.2 (OpenAlex 2-year mean citedness, value as of 2026-10-04, retrieved 2026-10-05; not the official Clarivate IF)Reference: Liu JK, Gu YQ, Sun QA, Chang JH, Zhao J, Liu YB, et al. Multiparametric MRI habitat radiomics and longitudinal dynamic clinical parameters predict TACE refractoriness in HCC. Sci Rep. 2026 Oct 9 [Epub ahead of print]. doi:10.1038/s41598-026-74555-8.Checked: Abstract only - Abstract
Anatomical liver resection for hepatocellular carcinoma (HCC) relies heavily on accurate intraoperative identification of portal venous anatomy. Although intraoperative ultrasound (IOUS) is regarded as the gold standard for real-time vascular navigation during liver surgery, identification of segment-specific portal vein branches remains technically demanding and highly dependent on surgical experience. Artificial intelligence (AI)-assisted ultrasound interpretation may improve the efficiency and consistency of intraoperative vessel recognition; however, most previous studies have been limited to static image analysis without real-time surgical application. This study aimed to develop and evaluate a YOLOv5-based deep-learning pipeline for automatic recognition of selected portal vein branches during laparoscopic intraoperative ultrasound and to investigate the feasibility of real-time intraoperative deployment during anatomical liver resection. A total of 3254 laparoscopic intraoperative ultrasound images obtained from 100 consecutive patients undergoing anatomical liver resection for HCC were retrospectively collected and manually annotated by experienced hepatobiliary surgeons. The annotated dataset was divided into independent training, validation, and testing cohorts (80%, 15%, and 5%, respectively) at the patient level. The final YOLOv5 model was trained using five predefined right-sided portal vein classes: P58, P67, P6, P7, and P8d. The same five-class model was used without retraining or class modification during prospective real-time validation in 20 patients. The model achieved a mean average precision ([email protected]) of 0.941 on the independent testing dataset. The maximum F1-score was 0.58 at a confidence threshold of 0.202, while precision reached 1.00 at a confidence threshold of 0.764. Branch-specific frame-level accuracy was 0.81 for P58, 0.81 for P67, 0.80 for P8d, 0.79 for P7, and 0.78 for P6. During prospective intraoperative validation, the system performed continuous real-time recognition with a processing latency of less than 50 ms per frame. The proposed pipeline demonstrated the technical feasibility of real-time recognition of five selected right-sided portal vein classes during laparoscopic intraoperative ultrasound. Further multicenter studies with broader anatomical coverage and larger prospective cohorts are required to establish generalizability and clinical utility.
Journal IF-equivalent: 3.5 (OpenAlex 2-year mean citedness, value as of 2026-10-10, retrieved 2026-10-11; not the official Clarivate IF)Reference: Xu J, Zhang X, Li S, Wei Y, Huang S, Cao L, et al. A deep-learning pipeline to assist portal vein identification during laparoscopic ultrasound scanning for anatomical liver resection in real time. Eur J Med Res. 2026 Oct 9 [Epub ahead of print]. doi:10.1186/s40001-026-05291-y.Checked: Abstract only
Kidney cancer: treatment, trials and AI
No new qualifying paper for this issue.
Bladder cancer: treatment, trials and AI
No new qualifying paper for this issue.