AI and Cancer Research ── 2026-10-19

9 papers in this issue. Lead: A comparative study of survival prediction and feature selection/extraction methods for quantitative radiomics risk modelling

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（1） · Lung（2） · Pancreatic（1） · Prostate（1） · ACC（0） · Brain（1） · Ovarian（0） · Endometrial（0） · Gastric（0） · Liver（1） · Kidney（0） · Bladder（0）

Cancer molecular biology and AI

A comparative study of survival prediction and feature selection/extraction methods for quantitative radiomics risk modelling
Heliyon · 2026-10-17
Abstract
Radiomics extracts quantitative imaging features from medical images to develop risk prediction models that support clinical decision-making. Owing to the high dimensionality and potential instability of radiomic features, accurate survival prediction depends critically on the selection of appropriate feature engineering and survival modelling methods. However, comprehensive benchmark studies comparing these methods remain limited, particularly for recently developed deep learning models. Here, we systematically benchmarked 16 survival prediction methods and 12 feature selection/extraction methods using three groups of publicly available radiomics datasets from non-small cell lung cancer and ovarian cancer cohorts, comprising 1160 patients and representing different levels of feature stability. Model performance was evaluated using repeated random training/testing splits and independent external validation with the C-index, integrated area under the curve (IAUC), and integrated Brier score (IBS). Ridge-Cox, glmboost, CoxBoost, and sslasso were consistently ranked among the top-performing methods according to the C-index and IAUC, whereas deep learning models, including CoxTime, DeepHit, and PCHazard, performed better when evaluated by IBS, highlighting that different evaluation metrics capture complementary aspects of survival prediction performance. Among feature engineering methods, supervised principal component analysis achieved the best overall performance, particularly for datasets with unstable radiomic features, whereas the Cox proportional hazards p -value-based method may be preferred when discrimination performance, as measured by the C-index or IAUC, is the primary objective. Furthermore, we investigated factors affecting the performance of deep learning models and assessed the influence of the proportional hazards and linear assumptions on survival prediction. These benchmark results provide practical guidance for selecting feature engineering and survival modelling strategies according to the characteristics of radiomics datasets, thereby facilitating the development of more robust radiomics-based survival prediction models.
Journal IF-equivalent: 9.5 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)
Reference: Liu W, Wang W. A comparative study of survival prediction and feature selection/extraction methods for quantitative radiomics risk modelling. Heliyon. 2026 Nov;12(16):e45602. doi:10.1016/j.heliyon.2026.e45602.
Checked: Abstract only
Brain foundation models with hypergraph dynamic adapter for brain disease analysis.
Pattern recognition ·
Abstract
Brain diseases, such as Alzheimer's disease and brain tumors, present profound challenges due to their complexity and societal impact. Recent advancements in brain foundation models have shown significant promise in addressing a range of brain-related tasks. However, current brain foundation models are limited by task and data homogeneity, restricted generalization beyond segmentation or classification, and inefficient adaptation to diverse clinical tasks. In this work, we propose SAM-Brain3D, a brain-specific foundation model trained on over 66,000 brain image-label pairs across 14 MRI sub-modalities, and Hypergraph Dynamic Adapter (HyDA), a lightweight adapter for efficient and effective downstream adaptation. SAM-Brain3D captures detailed brain-specific anatomical and modality priors for segmenting diverse brain targets and broader downstream tasks. HyDA leverages hypergraphs to fuse complementary multi-modal data and dynamically generate patient-specific convolutional kernels for multi-scale feature fusion and personalized patient-wise adaptation. Together, our framework excels across a broad spectrum of brain disease segmentation and classification tasks. Extensive experiments demonstrate that our method consistently outperforms existing state-of-the-art approaches, offering a new paradigm for brain disease analysis through multi-modal, multi-scale, and dynamic foundation modeling.
Journal IF-equivalent: 7.0 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)
Reference: Deng Z, Wang H, Huang Z, Zhang L, Aviles-Rivero AI, Liu C, et al. Brain foundation models with hypergraph dynamic adapter for brain disease analysis. Pattern Recognit. 2026 Apr;172:112595. doi:10.1016/j.patcog.2025.112595; PMCID: PMC7619582.
Checked: Full text checked

See all 2 Cancer biology papers →

Breast cancer: treatment, trials and AI

Explainable Radiogenomic Fusion Links Mri-derived Radiomic Phenotypes to Transcriptomic Molecular Subtypes in Breast Cancer
Research Journal of Life Sciences Bioinformatics Pharmaceutical and Chemical Sciences · 2026-10-17
Abstract
Radiogenomics aims to use non-invasive medical images to provide information about the molecular characteristics of tumors. However, many radiogenomic studies focus mainly on prediction accuracy and provide limited explanation of which imaging features drive the predictions or how those features relate to tumor biology. We developed an explainable radiogenomic framework for breast cancer that (1) compares machine-learning models for predicting molecular subtype from MRI radiomic features, (2) uses bootstrap permutation importance to identify a compact and stable radiomic signature, and (3) examines correlations between the selected features and gene expression. We first evaluated the workflow in a synthetic paired radiomics-transcriptomics cohort of 300 samples. Twelve MRI radiomic features and ten gene-expression values were generated from shared estrogen-receptor, HER2, and proliferation activity scores, providing a known ground truth for evaluation. Five-fold cross-validation showed strong four-class discrimination, with logistic regression achieving a macro-average AUC of 0.963. Bootstrap analysis using 100 resamples and a 70% stability threshold identified six robust features: margin spiculation, enhancement heterogeneity, wash-out ratio, shape irregularity, wash-in slope, and peak enhancement. The six-feature signature preserved classification performance (AUC = 0.966). Enhancement heterogeneity and margin spiculation were positively correlated with proliferation and basal markers and negatively correlated with estrogen-pathway genes, whereas wash-out ratio showed its strongest correlations with ERBB2 and GRB7. The synthetic cohort provides a controlled proof of concept showing that the workflow can recover predefined imaging-genomic relationships when a shared biological signal is present. These associations should not be interpreted as newly discovered patient-level biological findings. Independent validation using real paired breast MRI and transcriptomic cohorts is required to establish generalizability and clinical utility.
Journal IF-equivalent: 0.1 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)
Reference: Thakar SB, Thakar PS, Sangle A. Explainable Radiogenomic Fusion Links Mri-derived Radiomic Phenotypes to Transcriptomic Molecular Subtypes in Breast Cancer. Res J Life Sci Bioinform Pharm Chem Sci. 2026 Oct 17;12(5). doi:10.26479/2026.1205.02.
Checked: Abstract only

See all 1 Breast papers →

Lung cancer: treatment, trials and AI

THINKERS-X: multisequence MRI radiomics in a mixture-of-experts model for outcome prediction and exploratory dose-scenario evaluation in lung cancer brain metastases treated with radiosurgery.
Clinical & experimental metastasis · 2026-10-17
Abstract
We developed THINKERS-X: Lung, the eXtended multimodal evolution of THINKERS-Lung, to determine whether multisequence MRI radiomics provides incremental value for lesion-specific outcome prediction and exploratory dose-scenario evaluation after Gamma Knife radiosurgery (GKRS) for lung cancer brain metastases. We performed a retrospective single-center study of 767 patients with 3,728 lung cancer brain metastases treated with GKRS. THINKERS-X preserved the original mixture-of-experts and discrete-time survival architecture while adding a dedicated computational neuroradiology expert. Approximately 2,400 radiomic features per metastasis were extracted from pretreatment FSPGR/T1-weighted postcontrast, T2-weighted, and SWAN MRI. Validation used patient-grouped 5-fold cross-validation and a completely closed patient-grouped internal holdout cohort. In grouped cross-validation, THINKERS-X achieved a mean AUC 12 of 0.919 ± 0.021 versus 0.876 for THINKERS-Lung (paired difference, 0.043; 95% CI 0.026-0.061; p 12 improved from 0.863 to 0.929 (95% CI 0.895-0.958), with a paired difference of 0.066 (95% CI 0.031-0.101; p < 0.001). Mean absolute error decreased from 1.26 to 0.84 months. The calibrated Brier score was 0.039, calibration slope 0.94, and intercept 0.04. Radiomics altered model-derived dose-response profiles in 30.0% of metastases and the model-ranked candidate dose in 23.9%. THINKERS-X improved lesion-specific longitudinal prediction, temporal accuracy, and probabilistic calibration beyond THINKERS-Lung while materially altering exploratory dose-scenario evaluation in a subset of metastases. External multicenter and prospective validation are required before clinical implementation.
Journal IF-equivalent: 4.1 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)
Reference: Reyes JS, Hadjipanayis CG, Niranjan A. THINKERS-X: multisequence MRI radiomics in a mixture-of-experts model for outcome prediction and exploratory dose-scenario evaluation in lung cancer brain metastases treated with radiosurgery. Clin Exp Metastasis. 2026 Oct 17;43(6):59. doi:10.1007/s10585-026-10443-9.
Checked: Abstract only
AI-Based Cancer Drug Response Prediction: A Machine Learning Ensemble for Lung Cancer Pharmacogenomic Classification
International Journal of Scientific Research in Computer Science Engineering and Information Technology · 2026-10-17
Abstract
Effectiveness of anticancer drug varies among various types of cancers so understanding their response towards these medications is essential. The paper describes an algorithm to classify drug response patterns from the lung cancer subset of GDSC2 data set through machine learning techniques. Continuous LN_IC50 response values are classified as sensitive and resistant based on predetermined cut-off points for creating a balanced dataset for modeling purposes. Eight out of nine numbers are obtained by drugs, cells lines targets pathways related responses while selecting them via selectkbest method through ANOVA Ftest. Logistic Regression, Support Vector Machine, Random Forest and XGBoost models were trained and evaluated on a held-out test set of 6,404 samples from an unbalanced dataset containing 32,018 records. Test accuracy of these models was between 92.61% and 92.96%, whereas their ROC-AUC value ranged from 0.978 to 0.981. Integrated with a Django based research tool by means of an algorithm that combines predictions via vote for producing outputs from this system. This device is meant solely to serve as an educational tool; it cannot be used by patients for self-diagnosis, healthcare decisions, or prescription of medications.
Journal IF-equivalent: 1.0 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)
Reference: P KP, Mohanapriya D, Thara L. AI-Based Cancer Drug Response Prediction: A Machine Learning Ensemble for Lung Cancer Pharmacogenomic Classification. Int J Sci Res Comput Sci Eng Inf Technol. 2026 Oct 17;12(5):466-78. doi:10.32628/cseit2612553.
Checked: Abstract only

See all 2 Lung papers →

Pancreatic cancer: treatment, trials and AI

Deep-learning–based automated multimodal fusion network using multiphoton imaging for predicting lymph-node metastasis in pancreatic ductal adenocarcinoma
Biomedical Signal Processing and Control · 2026-10-17
Abstract
Lymph-node metastasis is a key factor that influences clinical outcomes in patients with pancreatic ductal adenocarcinoma (PDAC), and its accurate and timely identification is essential for optimizing therapeutic strategies and improving patient prognosis. However, the need for extensive screening of pathological scans makes its diagnosis challenging. Multiphoton microscopy (MPM) is a powerful tool for label-free, two- and three-dimensional visualizations of tissue, which exhibits considerable promise in pancreatic-tumor diagnostics. The automation of diagnostic processes has substantially alleviated the workload of pathologists, while maintaining high accuracy. This study aims to develop an efficient deep-learning framework for the automatic prediction of lymph-node metastasis in PDAC patients, using MPM images. We evaluate the performance of four deep-learning models: VGG16, ResNet50, MobileNetv2, and pyramid vision transformer version 2 (PVTv2). Based on the superior and stable performance of PVTv2, we develop a combined model, named automatic lymph-node–metastasis classification network (AutoLNMClassNet), which integrates PVTv2-derived features with collagen and clinical features through a multilayer perceptron. The proposed-model’s performance is evaluated in terms of the accuracy, positive/negative predictive values, sensitivity, and specificity. AutoLNMClassNet demonstrates superior performance, achieving an accuracy of 0.922 and area under the curve of 0.969, outperforming other classification models. These results indicate that AutoLNMClassNet offers high reliability and accuracy in detecting lymph-node metastases, thereby serving as a valuable tool for the diagnosis and treatment of PDAC.
Journal IF-equivalent: 5.1 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)
Reference: Xi G, Huang Y, He J, Wang G, Zhang J, Chen L, et al. Deep-learning–based automated multimodal fusion network using multiphoton imaging for predicting lymph-node metastasis in pancreatic ductal adenocarcinoma. Biomedical Signal Processing and Control. 2027 Mar;131:111667. doi:10.1016/j.bspc.2026.111667.
Checked: Abstract only

See all 1 Pancreatic papers →