AI and Cancer Research ── 2026-11-07
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.
All cancers: Cancer biology（0）
Hematologic malignancies: AML（0） · CML（0） · ALL（0） · CLL（0） · Lymphoma（0） · Myeloma（0）
Solid tumors: Breast（1） · Lung（2） · Colorectal（2） · Pancreatic（0） · Prostate（1） · Brain（2） · Ovarian（0） · Endometrial（0） · Gastric（1） · Liver（1） · Kidney（1） · Bladder（1）
Rare cancers: ACC（0）
Cancer molecular biology and AI
No new qualifying paper for this issue.
Hematologic malignancies
Acute myeloid leukemia (AML): treatment, trials and AI
No new qualifying paper for this issue.
Chronic myeloid leukemia (CML): treatment, trials and AI
No new qualifying paper for this issue.
Acute lymphoblastic leukemia (ALL): treatment, trials and AI
No new qualifying paper for this issue.
Chronic lymphocytic leukemia (CLL): treatment, trials and AI
No new qualifying paper for this issue.
Lymphoma: treatment, trials and AI
No new qualifying paper for this issue.
Multiple myeloma: treatment, trials and AI
No new qualifying paper for this issue.
Solid tumors
Breast cancer: treatment, trials and AI
- Abstract
Background: Breast cancer treatment depends on histopathological features, such as grade and receptor-defined subtype; however, specialist pathologist access is constrained when the workforce is limited. Commercial multimodal large language models (MLLMs) accept hematoxylin and eosin (H&E) image tiles through paid interfaces without local hardware or fine-tuning. However, prior pathology evaluations addressed only coarse tasks. Whether they reach treatment-determining accuracy and whether vendors agree remain unclear.
Methods: We aimed to evaluate three vendor-designated flagship MLLMs (Claude Sonnet 4.6, Gemini 2.5 Pro, GPT-5.5) in 427 invasive breast cancer cases. Each case went to all three with identical H&E tiles and prompts, and the subtype was inferred in the second call. The reference was an institutional sign-out report of an immunohistochemistry-derived subtype. We calculated the concordance, sensitivity, specificity, Cohen’s kappa, and pairwise McNemar and Bowker tests.
Findings: Claude ranked highest by raw histologic-type concordance but lowest by kappa, classifying all 23 lobular and seven micropapillary carcinomas as invasive breast carcinoma of no special type. The models anchored the Nottingham grade to three modal grades. None of the models reliably identified human epidermal growth factor receptor 2-positive disease. The failure direction was vendor-specific: Claude and GPT-5.5 were under-detected, whereas Gemini was over-called. Twelve prompt variants (4,056 calls) did not recover sensitivity.
Interpretation: No current commercial MLLM reaches deployment-ready accuracy for any treatment-determining feature of breast pathology. As each vendor fails in its own fixed direction, changing vendors alters the type of error rather than removing it; therefore, the value of these models is assistive rather than autonomous. At USD 0.20–0.50 per case, they may serve as supervised draft generators that leave the diagnosis with the pathologist. Research in context Evidence before this study We searched PubMed and Embase from database inception to January 10, 2026, without language restriction, using combinations of the terms “large language model,” “multimodal,” “GPT,” “Gemini,” “Claude,” “foundation model,” “breast cancer,” “pathology,” “histopathology,” “whole-slide image,” and “diagnostic accuracy.” We also screened the reference lists of retrieved articles. Task-specific deep-learning models and pathology foundation models pretrained on large slide collections achieve strong performance on individual breast pathology tasks such as Nottingham grading and receptor-status prediction, but require local GPU infrastructure, curated training data, and deployment expertise. Evaluations of general-purpose commercially available multimodal large language models (MLLMs) in pathology were limited to coarse tasks, such as tissue-type classification and metastasis, and were typically confined to a single model. We found no studies directly comparing current flagship commercial MLLMs on clinically relevant, treatment-determining breast pathology tasks, and none reporting whether vendors agree or whether the choice of vendor changes the diagnosis. The available evidence was limited by single-model designs, task-narrow datasets, and reliance on raw accuracy without chance-corrected agreement. Added value of this study In this paired, single-center retrospective study of 427 invasive breast cancers, we evaluated three vendor-designated flagship commercial MLLMs (Claude Sonnet 4.6, Gemini 2.5 Pro, GPT-5.5) on identical H&E tiles and prompts against an expert pathologist reference, across treatment-determining features. No model reached deployment-ready accuracy for any feature. Raw concordance and chance-corrected agreement ranked the models in opposite order, so a decision based on raw accuracy alone would have selected the least discriminating model. Each model demonstrated a consistent error pattern, tending either toward under-detection or over-call. Consequently, changing vendors altered the type of diagnostic error rather than removing it. Twelve prompt variants across 4,056 calls, a reasoning-effort escalation, and an intra-vendor version upgrade did not change the failure direction, indicating a vendor-specific prior rather than a prompt-engineering artifact. To our knowledge, this is the first head-to-head, chance-corrected comparison of commercial MLLMs at the level of treatment-determining breast pathology. Implications of all the available evidence Taken together with previous evidence, our findings indicate that current commercial MLLMs are not ready for autonomous interpretation of breast pathology and should not be used as primary readers without pathologist oversight. Their value, if any, is assistive. Given their relatively low per-case cost, these systems may be useful for generating supervised draft reports where pathologist workload is the primary constraint, provided that digital pathology infrastructure and immunohistochemical testing are available and that final diagnostic responsibility remains with a qualified pathologist. Because the failure direction is vendor-specific, deployment requires vendor-aware caveats and item-level, chance-corrected evaluation rather than raw accuracy. Improving performance will likely require incorporation of domain-specific knowledge through approaches such as in-context reference examples, retrieval against a curated atlas, or handoff to a domain-trained model. Future development should be supported by immunohistochemistry-grounded datasets and prospective validation in the target setting.
Journal IF-equivalent: 5.8 (OpenAlex 2-year mean citedness, value as of 2026-10-21, retrieved 2026-10-22; not the official Clarivate IF)Reference: Jun SY, Kim S, An S, Kang YJ. Commercial multimodal large language models show systematic model-specific errors in breast cancer pathology reporting: a paired concordance study. Breast Cancer Res. 2026 Nov 4 [Epub ahead of print]. doi:10.1186/s13058-026-02413-6.Checked: Abstract only
Lung cancer: treatment, trials and AI
- Abstract
Introduction: The 2020 International Association for the Study of Lung Cancer (IASLC) grading system integrates the predominant histologic pattern and the proportion of high-grade patterns, but preoperative CT assessment is limited by spatial averaging of heterogeneous tumor components. We aimed to develop a CT-derived nested habitats score (NHS) that recursively identifies grade-associated intratumoral CT subregions, evaluate its cross-center performance for preoperative IASLC grading, and explore its transcriptomic correlates.
Methods: This retrospective two-center study included 650 patients with surgically resected lung adenocarcinoma. Center 1 formed the development cohort (n = 390); Center 2 was stratified 1:1 into validation (n = 130) and locked test (n = 130) cohorts, both from the same external institution. The prespecified primary task was pathologic IASLC Grade 1 versus Grade 2/3. NHS was generated using five-fold out-of-fold estimation in development data. Clinical-radiological, whole-tumor radiomics, NHS, and integrated models were compared, with algorithm selection confined to validation before one-time test evaluation. An independent TCGA-LUAD/TCIA CT-RNA cohort (n = 88) underwent exploratory WGCNA and Hallmark gene-set enrichment analysis.
Results: The development/validation/test cohorts contained 41/14/14 Grade 1, 293/101/96 Grade 2, and 56/15/20 Grade 3 tumors. The selected random-forest integrated model yielded AUCs of 0.871 (95% CI, 0.784–0.936) in validation and 0.866 (95% CI, 0.784–0.935) in testing. NHS alone yielded AUCs of 0.828 and 0.854, compared with 0.620 and 0.497 for the clinical-radiological model and 0.703 and 0.775 for whole-tumor radiomics. NHS accounted for 76.3% of mean absolute SHAP contribution; removing NHS reduced AUC by 0.251 (95% CI, 0.079–0.439) in validation and 0.369 (95% CI, 0.200–0.553) in testing. Brier scores for NHS and the integrated model (0.074–0.077) were below the non-informative reference (0.096), but calibration slopes exceeded 1, and decision curves showed no net benefit over treating all patients as Grade 2/3 at low-to-intermediate thresholds. Exploratory radiogenomics showed nominal NHS-associated coexpression modules and enrichment of WNT/beta-catenin, E2F, G2M-checkpoint, and immune-related programs. Discussion NHS discriminated IASLC Grade 1 from Grade 2/3 tumors, retained stable performance in two held-out cohorts from a second institution, and contributed most of the integrated model’s discrimination. The transcriptomic associations are hypothesis-generating, and recalibration and prospective multicenter evaluation are required before clinical application.
Journal IF-equivalent: 4.8 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Peng W, Liu J, Li L, Qi W, Zhao J, Li D. CT-derived nested habitats score for preoperative IASLC grading in lung adenocarcinoma: cross-center validation and radiogenomic analysis. Front Mol Biosci. 2026 Nov 4;13:2006861. doi:10.3389/fmolb.2026.2006861.Checked: Abstract only - Abstract
Phage display can identify ligands that target distinctive cell surface phenotypes, but selection outputs are often complex and difficult to interpret. Here, we developed an integrated workflow combining phage display biopanning, next-generation sequencing (NGS), machine learning (ML), and flow cytometry to discover short peptide binders associated with genotype-defined cancer contexts. Phage display was performed on lung cancer cell lines representing KRAS wild-type (WT), G12C, G12D, and G12V backgrounds, and experimental outputs were used to train a multimodal probabilistic model. From the resulting model-generated candidates, 98 nonredundant heptapeptides were selected for synthesis, and 88 were successfully solubilized and tested experimentally. Flow cytometry validation revealed heterogeneous binding behaviors, including broad pan-cellular binders and candidates with mutant-vs-WT binding preference. Extended dose–response profiling and normal-cell counter-screening enabled prioritization based on binding magnitude, curve behavior, normal-cell reactivity, and sequence developability. Overall, this study shows how whole-cell selection data sets can be analyzed computationally and experimentally to identify and prioritize peptide candidates for targeting strategies.
Journal IF-equivalent: 5.2 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Uguzzoni G, Belisario DC, Pagnani A, Marchiò S. Integrated Whole-Cell Phage Display and Machine Learning Identify Candidate Targeting Ligands in KRAS-Defined Lung Cancer Contexts. ACS Omega. 2026 Nov 4 [Epub ahead of print]. doi:10.1021/acsomega.6c10403.Checked: Abstract only
Colorectal cancer: screening, treatment, trials and AI
- Abstract
Background: Deep-learning computational pathology applied to routine haematoxylin and eosin whole-slide images may provide prognostic information in colorectal cancer. This systematic review evaluated the association between image-derived risk groups and survival.
Methods: PubMed and Web of Science Core Collection were searched from inception to 21 June 2026. Studies of colorectal cancer reporting a high-versus-low-risk hazard ratio derived from haematoxylin and eosin images were included. The most-adjusted estimate was selected, with preference for external validation. Random-effects meta-analysis, subgroup and leave-one-out analyses were performed. The recovered risk-of-bias panel reports Quality in Prognosis Studies domain judgements. The review was registered in PROSPERO (CRD420261429467).
Results: Across 27 studies, the pooled hazard ratio was 2.49 (95% confidence interval 2.12–2.93), with substantial heterogeneity (I² = 77%) and a 95% prediction interval of 1.25–4.97. Pooled estimates were 2.36 in externally validated studies (17 studies) and 2.74 in internally validated studies (10 studies); the difference was not statistically significant. Leave-one-out estimates ranged from 2.35 to 2.57.
Conclusions: Image-derived risk scores were associated with survival in colorectal cancer. Heterogeneity, differences in risk definitions and possible overlap between validation cohorts limit interpretation of a single pooled estimate. Independent prospective validation and evaluation of clinical utility are needed before these methods can guide treatment decisions.
Journal IF: unknown (could not be matched)Reference: Xu Y, Zhang L, Wang C, Yao Y, Shen C, Gu W. Prognostic value of deep-learning computational pathology in colorectal cancer: a systematic review of haematoxylin and eosin whole-slide image studies. Research Square. 2026 Nov 4 [Epub ahead of print]. doi:10.21203/rs.3.rs-11297991/v1.Checked: Abstract only - Abstract
BACKGROUND: Severe postoperative complications after rectal cancer surgery increase morbidity, delay recovery, and may compromise oncologic outcomes. Existing risk models show limited accuracy and generalizability. Machine learning can integrate complex clinical and surgical variables to improve individualized risk prediction. OBJECTIVE: To develop and prospectively validate machine learning-based models for predicting severe postoperative complications after rectal cancer surgery. DESIGN: Retrospective model development and prospective temporal internal validation study. SETTING: Single tertiary referral center in South Korea. PATIENTS: A total of 3,503 patients who underwent curative rectal cancer resection with anastomosis between 2011 and 2025 were included, comprising 3,402 retrospective and 101 prospective cases. MAIN OUTCOME MEASURES: Severe postoperative complications defined as Clavien-Dindo grade ≥ 3 within 30days after surgery. RESULTS: Severe complications occurred in 9.8% of the retrospective and 13.9% of the prospective cohort. Four machine-learning algorithms - logistic regression, random forest, artificial neural network, and gradient boosting - were trained using 5-fold crossvalidation. In internal validation, the random forest model achieved the best performance (area under the receiver operating characteristic Curve 0.898; Sensitivity 0.685), while gradient boosting performed best in the temporal internal validation cohort (area under the receiver operating characteristic Curve 0.910; Sensitivity 0.707). The most influential predictors were tumor perforation, intraoperative transfusion, protective stoma, bowel obstruction, and cerebrovascular accident. LIMITATIONS: Validation was single institution (temporal internal, not external) with a small prospective cohort. CONCLUSIONS: Machine learning-based models accurately predicted severe postoperative complications after rectal cancer surgery, with strong discrimination maintained on temporal internal validation. Applied immediately after surgery, they may support individualized postoperative risk stratification and monitoring; external validation is warranted. See Video Abstract.
Journal IF-equivalent: 1.7 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Song SH, Kwon HK, Park JS, Park SY, Kim HJ, Lee J, et al. Machine Learning-Based Risk Scoring Using Clinical and Surgical Variables for Severe Complications after Rectal Cancer Surgery. Diseases of the Colon & Rectum. 2026 Nov 4 [Epub ahead of print]. doi:10.1097/dcr.0000000000004526.Checked: Abstract only
Pancreatic cancer: treatment, trials and AI
No new qualifying paper for this issue.
Prostate cancer: treatment, trials and AI
- Abstract
Aim: Large language models (LLMs) may help interpret clinical guideline content. However, their performance in guideline-based prostate cancer (PC) scenarios remains insufficiently validated. This study evaluated the agreement between a multidisciplinary expert panel and responses generated by three leading LLMs (Claude, Gemini and ChatGPT) when addressing structured, guideline-derived clinical questions on PC. Materials & methods: A multidisciplinary tumor board created 25 structured questions based on the 2025 European Association of Urology guidelines. The guideline figures and tables were transformed into plain-language summaries for use by the LLMs. Each model was prompted with a simple, unmodified query and a panel of experts in PC management independently rated their level of agreement with the responses. The evaluation used a two-round modified Delphi method with predefined consensus criteria.
Results: Eleven experts evaluated 25 guideline-based questions. After two Delphi rounds, consensus in agreement was reached for 23/25 responses (92%) with Gemini, 20/25 (80%) with ChatGPT and 14/25 (56%) with Claude (p = 0.0033). Explicit disagreement was noted in three questions (12%) with Claude after the second round. Median scores of at least 7 out of 9, indicating strong agreement, were achieved in 24/25 responses (96%) for Gemini, 22/25 (88%) for ChatGPT, and 14/25 (56%) for Claude (p = 0.0002). Most consensus difficulties involved complex cases, advanced imaging, genetic testing or questions with limited data or controversial topics.
Conclusion: Advanced LLMs, when provided with PC guidelines with plain-language adaptations of figures and tables, can achieve strong alignment with expert consensus in routine situations. However, performance varies among models, with Claude showing lower agreement rates, especially in complex cases. These results emphasize the importance of expert oversight, continuous validation and standardized guideline formats interpretable by both clinicians and AI systems. What is this article about? Clinical guidelines help doctors make decisions about prostate cancer (PC), but they can be long and complex. Artificial intelligence tools such as Claude, Gemini and ChatGPT may help interpret guideline information. However, the answers generated by these tools need to be checked by medical experts before they can be used in practice. What methodology was used in this study? PC specialists created 25 questions based on European PC guidelines. The original guideline text was provided to the three artificial intelligence tools, with figures and tables converted into clear, structured text. For each question, experts reviewed the answers and rated whether they agreed with them. What were the results? Gemini and ChatGPT agreed with expert opinion more often than Claude. After two rounds of expert review, Gemini reached agreement in 23 of 25 answers, ChatGPT in 20 of 25 answers, and Claude in 14 of 25 answers. The tools performed better for routine PC questions than for more complex situations, such as advanced imaging, genetic testing, incomplete clinical information or controversial topics. Why are these results important? These results suggest that artificial intelligence tools may help interpret information from PC guidelines, especially in straightforward situations. However, their responses are not always reliable and should be checked by healthcare professionals with expertise in PC.
Journal IF-equivalent: 2.7 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Borque-Fernando Á, Ledo Cepero MJ, Rodríguez-Fernández A, Espinós EL, Breijo SM, Gordoa TA, et al. Evaluation of large language models for clinical decision support in prostate cancer: a multidisciplinary Delphi study. J Comp Eff Res. 2026 Nov 4 [Epub ahead of print]. doi:10.57264/cer-2026-0081.Checked: Abstract only
Brain tumours: treatment, trials and AI
- Abstract
Objective: To evaluate contrast-enhanced T1-weighted MRI (CE-T1WI) models for preoperative prediction of isocitrate dehydrogenase (IDH) mutation status in adult-type diffuse gliomas and explore phenotype associations of influential latent features.
Methods: This retrospective study included 462 patients. Center 1 provided training ( n = 259) and internal-test ( n = 65) cohorts, and Center 2 provided external validation ( n = 138). Five input–architecture combinations and six classifiers were evaluated against an interpretable logistic benchmark combining age and Visually Accessible Rembrandt Images (VASARI) descriptors. For each architecture, the classifier with the highest observed external area under the curve (AUC) was selected for reporting. Performance was assessed using discrimination, calibration, decision curves, and 100 downstream refits with fixed deep features. Shapley additive explanations (SHAP) and phenotype association analyses examined influential latent features and their relationships with clinical and imaging phenotypes.
Results: Region-of-interest (ROI)-constrained three-dimensional vision Transformer (3D ViT) with a support vector machine (SVM) achieved an external AUC of 0.929 (95% CI, 0.885–0.973), accuracy of 0.848, sensitivity of 0.838, specificity of 0.857, and Brier score of 0.106. The age–VASARI benchmark achieved an external AUC of 0.782. Across downstream refits, the ROI-constrained model had a mean external AUC of 0.930 and the smallest mean training-minus-external AUC difference (0.019). The same 10 features formed the highest-importance SHAP set across cohorts, with different rank orders. Feature_546 correlated with age ( r = 0.461), postoperative Ki-67 ( r = 0.562), and enhancement quality ( r = 0.678; all adjusted p < 0.001).
Conclusion: ROI-constrained volumetric learning from CE-T1WI showed promising cross-center IDH discrimination using one routine MRI sequence. Influential latent features were associated with postoperative Ki-67 and recognizable imaging phenotypes. SHAP attribution and these exploratory associations provide complementary perspectives on model behavior in preoperative IDH assessment.
Journal IF-equivalent: 3.4 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Zheng C, Zhang Q, Li D, Wang H, Yang J, Lv F, et al. Explainable volumetric deep learning from contrast-enhanced T1-weighted MRI for preoperative IDH prediction in adult-type diffuse gliomas: a two-center study. Front Neurol. 2026 Nov 4;17:1966591. doi:10.3389/fneur.2026.1966591.Checked: Abstract only - Abstract
OBJECTIVE: To develop ML models predicting postoperative CSF leakage after endoscopic transsphenoidal pituitary adenectomy (TSPA). METHODS: Four hundred and twenty patients from five hospitals (January 2021-December 2022) comprised the training cohort; 140 from another hospital were the validation cohort. The training cohort included 60 CSF leak and 360 nonleak cases. Risk factors were analyzed. Decision tree, logistic regression, and SMOTE-based models were built and their diagnostic value compared. RESULTS: Significant differences ( p < 0.05) existed in tumor diameter, tumor texture, operative time, intraoperative diaphragma sellae rupture, reoperation, and intraoperative CSF leakage grade. Logistic regression identified tumor texture, reoperation, intraoperative CSF leakage grade, and diaphragma sellae rupture as risk factors. All models had high AUCs in training: logistic regression (0.792), decision tree (0.835), and SMOTE (0.839). The Delong test showed a decision tree, and SMOTE outperformed logistic regression. SMOTE had the highest sensitivity/specificity. Validation cohort AUCs were: logistic regression (0.775), decision tree (0.809), and SMOTE (0.831), with SMOTE superior. Calibration curves indicated good agreement between predicted and actual values. CONCLUSION: Prediction models for CSF leakage postendoscopic TSPA were developed using decision trees, SMOTE, and logistic regression. Model efficacy was evaluated via AUC, specificity, and sensitivity. Comparing algorithms reduced single-algorithm bias, with SMOTE demonstrating the best performance.
Journal IF-equivalent: 0.1 (OpenAlex 2-year mean citedness, value as of 2026-10-29, retrieved 2026-10-31; not the official Clarivate IF)Reference: Shao S, Ye C, Du Y, Liu T, Wu Y, Chen M, et al. Construction of a Predictive Model for Cerebrospinal Fluid Leakage After Endoscopic Transsphenoidal Pituitary Tumor Surgery Based on Machine Learning Algorithm. J Neurol Surg B Skull Base. 2026 Jan 6;87(06):669-77. doi:10.1055/a-2775-5314.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
- Abstract
Background/Objectives: Human epidermal growth factor receptor 2 (HER2) is an important biomarker in gastric cancer (GC). Histologic assessment may be limited by tissue sampling and spatial heterogeneity. We evaluated radiomics models for preoperative HER2 classification.
Methods: PubMed, the Cochrane Library, Embase, and Web of Science were searched from inception to 7 November 2024 and updated through 10 August 2026. A Study_ID–Report_ID–Cohort_ID index identified overlaps. AUROC (C-statistic) values were pooled on the logit scale using random-effects models. Patient-level diagnostic accuracy was pooled using bivariate random-effects models only for directly reported 2 × 2 tables or unique integer reconstructions. Quality and reporting were assessed using RQS, PROBAST+AI, and TRIPOD+AI.
Results: Twenty-three eligible reports represented 20 independent studies and three overlapping companion reports. The modeling population comprised 6,734 patients, including 1,484 with HER2-positive tumors. Pooled AUROC was 0.812 (95% CI, 0.763–0.853; I² = 78.9%) in training cohorts and 0.821 (95% CI, 0.772–0.861; I² = 46.3%) in validation cohorts. For validation CECT models, AUROC was 0.824 (95% CI, 0.772–0.866; I² = 55.5%). In the diagnostic accuracy meta-analysis, pooled sensitivity and specificity were 0.718 and 0.764 in five training cohorts and 0.714 and 0.827 in four validation cohorts.
Conclusions: Radiomics models provided preoperative discrimination of HER2 status. CECT has the largest evidence base and may serve as a research adjunct to histologic testing. PET/CT, DECT, and deep learning require further prospective multicenter validation, with standardized imaging and complete reporting of calibration, decision curves, and clinical utility.
Journal IF-equivalent: 2.3 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Ma H, Tang JW, Tang SZ, Zhu FZ, Yin JX, Zou SY, et al. Preoperative human epidermal growth factor receptor 2 assessment in gastric cancer using radiomics and machine learning: A diagnostic meta-analysis. World J Gastrointest Oncol. 2026 Nov 4 [Epub ahead of print]. doi:10.4251/wjgo.127294.Checked: Abstract only
Liver cancer: treatment, trials and AI
- Abstract
PURPOSE: Tumor heterogeneity hinders accurate prediction of response to immune checkpoint inhibitors (ICIs) in hepatocellular carcinoma (HCC). We developed a habitat model using routine in-phase (IP) and opposed-phase (OP) MRI sequences, which provide complementary structure- and lipid-related information, to predict response to ICI therapy in HCC. METHODS: This study included 883 HCC patients from six institutions and public databases. We developed habitat, radiomics, peritumoral, clinical, and combined models to predict partial response (PR) in five institutions. Kaplan-Meier analysis was performed to evaluate survival differences based on prediction scores. Habitat model interpretability was assessed using Shapley additive explanations (SHAP) along with multi-omics and pathological analyses in the public databases and one external institution. RESULTS: The habitat model showed numerically higher AUCs among the independent models, with AUCs of 0.923, 0.918, and 0.897 in the training set and two external validation sets, respectively. The combined model also showed favorable performance with AUCs of 0.959, 0.965, and 0.955, respectively. Calibration curves showed generally close agreement, while decision curve analyses suggested potential clinical net benefit of the habitat and combined models. Patients with higher habitat scores had significantly longer progression‑free and overall survival (all P < 0.05). Exploratory biological analyses showed that the higher habitat score group was associated with enhanced immune activity, downregulated fatty acid-related pathways, and shorter tumor-inflammatory spatial distances. CONCLUSION: The IP and OP MRI-based habitat model showed potential for treatment-response prediction and prognostic stratification in patients with HCC receiving ICI therapy.
Journal IF-equivalent: 4.8 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Fu X, Guo Y, Gong B, Chen X, Wang N, Hu H, et al. Decoding tumor heterogeneity using MRI for predicting response to immune checkpoint inhibitors in hepatocellular carcinoma: a multicenter study. Radiol med. 2026 Nov 4 [Epub ahead of print]. doi:10.1007/s11547-026-02303-z.Checked: Abstract only
Kidney cancer: treatment, trials and AI
- Abstract
Aim: Large language models (LLMs) may facilitate the interpretation of clinical guideline information, but their performance in renal cell carcinoma (RCC) remains insufficiently characterized. We evaluated agreement between a multidisciplinary expert panel and the responses generated by Claude, Gemini and ChatGPT to RCC guideline-based questions. Materials & methods: A multidisciplinary tumor board developed 25 clinical questions based on the 2025 European Association of Urology RCC guidelines. Guideline figures and tables were converted into structured, plain-language textual summaries and supplied to the models. Each LLM generated one response per question using direct prompting. An 11-member multidisciplinary expert panel evaluated the responses in a two-round modified Delphi process using a 9-point scale and predefined consensus criteria.
Results: Consensus in agreement was reached for 24/25 Gemini responses (96%) and 23/25 responses (92%) for both Claude and ChatGPT. No response reached consensus in disagreement. The proportion of responses with a median score ≥7 was 25/25 (100%) for Gemini, 24/25 (96%) for Claude and 23/25 (92%) for ChatGPT. No statistically significant differences were detected among the models overall or in pairwise comparisons. Responses requiring further discussion or failing to reach consensus mainly involved clinically uncertain or individualized scenarios. Panelists identified occasional factual inaccuracies, omitted options and limited clinical nuance.
Conclusion: When provided with structured, guideline-derived material, selected LLMs generated responses that frequently aligned with multidisciplinary expert consensus. These findings support their potential for guideline interpretation but do not demonstrate independent clinical reasoning or real-world decision-making performance. Expert oversight remains essential, particularly in complex scenarios. What is this article about? Clinical guidelines help doctors make decisions about renal cell carcinoma, the most common type of kidney cancer, but they can be long and complex. Artificial intelligence tools such as Claude, Gemini and ChatGPT may help interpret this information. However, their answers need to be checked by medical experts before they are used in practice. What methodology was used in this study? Specialists created 25 questions based on the European Association of Urology guidelines for kidney cancer. The original guideline text was provided to the three artificial intelligence tools, with figures and tables converted into clear, structured text. An expert panel of 11 specialists then reviewed the answers and rated whether they agreed with them in a two-round process. What were the results? The expert panel reached agreement with 96% of Gemini responses and 92% of responses from both Claude and ChatGPT. There were no clear overall differences between the three tools. Difficulties were more common in complex situations, such as careful monitoring without immediate treatment in kidney cancer that has spread, expected outcomes in a less common type of kidney cancer, choosing treatments that destroy the tumor without removing it surgically, and considering imaging tests as possible alternatives to kidney biopsy. Experts also identified some incorrect information, missing options and limited detail. Why are these results important? These tools may help interpret kidney cancer guideline information. However, they were tested using information supplied from the guidelines and were not shown to make independent clinical decisions. Their responses should therefore be checked by healthcare professionals with expertise in kidney cancer.
Journal IF-equivalent: 2.7 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Montes SFP, García IO, Martín RB, Agustín Ferrández MJ, Arrea FA, Esteban EC, et al. Evaluation of large language models for clinical decision support in renal cell carcinoma: a multidisciplinary Delphi study. J Comp Eff Res. 2026 Nov 4 [Epub ahead of print]. doi:10.57264/cer-2026-0082.Checked: Abstract only
Bladder cancer: treatment, trials and AI
- Abstract
Aim: To assess expert panel agreement with responses from three large language models, Claude, Gemini and ChatGPT, to questions based on guidelines for non-muscle-invasive bladder cancer (NMIBC), muscle-invasive bladder cancer (MIBC) and upper tract urothelial carcinoma (UTUC). Materials & methods: The 2025 European Association of Urology (EAU) guidelines were provided to Claude 3.5 Haiku, Gemini 2.5 Flash and GPT-4.1-mini. Figures and tables were converted into plain language text. Each model answered 25 questions per entity. Panels of 7-10 experts rated responses on a 9-point scale using a modified Delphi process with two rounds. Outcomes were consensus in agreement and median expert panel score ≥7.
Results: For Claude, Gemini and ChatGPT, respectively, consensus in agreement was achieved for 88%, 100% and 96% of NMIBC responses (p = 0.097); 92%, 100% and 96% of MIBC responses (p = 0.223); and 56%, 100% and 96% of UTUC responses (p Conclusion: Under the study conditions, large language model responses frequently achieved expert panel agreement, particularly for Gemini and ChatGPT, supporting the potential of these models for guideline interpretation. Further validation is required to establish clinical utility and safety. What is this article about? Clinical guidelines help healthcare professionals make decisions about bladder and upper urinary tract cancers, but they can be long and complex. Artificial intelligence tools such as Claude, Gemini and ChatGPT may help interpret guideline information. However, their answers need expert review before clinical use. What methodology was used in this study? Specialists created 25 questions for three clinical areas: non-muscle-invasive bladder cancer, muscle-invasive bladder cancer and upper tract urothelial carcinoma. The 2025 European Association of Urology guidelines were provided to the three tools, with figures and tables converted into clear text. Each tool answered 75 questions. Panels of 7–10 specialists reviewed the responses in two rounds. What were the results? Across all 75 questions, experts agreed with 59 Claude responses, 75 Gemini responses and 72 ChatGPT responses. The largest differences occurred in upper tract urothelial carcinoma, where experts agreed with 14 of 25 Claude responses, 25 of 25 Gemini responses and 24 of 25 ChatGPT responses. Expert comments highlighted limitations in the completeness and focus of some Claude responses. Why are these results important? These tools can often interpret supplied guideline information in a way that agrees with specialist opinion. However, the study did not assess their safety or usefulness in real clinical practice. Their responses should therefore be checked by healthcare professionals, particularly in complex situations.
Journal IF-equivalent: 2.7 (OpenAlex 2-year mean citedness, value as of 2026-11-05, retrieved 2026-11-06; not the official Clarivate IF)Reference: Guerrero-Ramos F, Fernández Gómez JM, Aparicio TF, López-Beltrán A, Rueda ÓB, González LL, et al. Evaluation of large language models for guideline-based clinical questions in muscle-invasive and non-muscle-invasive bladder cancer and upper tract urothelial carcinoma: a multidisciplinary Delphi study. J Comp Eff Res. 2026 Nov 4 [Epub ahead of print]. doi:10.57264/cer-2026-0079.Checked: Abstract only
Rare cancers
Adenoid cystic carcinoma: treatment, trials and AI
No new qualifying paper for this issue.