AI and Cancer Research ── Lung ── 2026-11-08
Lung cancer: treatment, trials and AI
- Abstract
Lung cancer clinical decision support demands precise reasoning across complex, multi-stage oncological workflows. Existing multimodal large language models (MLLMs) fail to handle guideline-constrained staging and treatment reasoning. We formalize three oncological precision treatment (OPT) tasks for lung cancer, spanning TNM staging, treatment recommendation, and end-to-end clinical decision support. We introduce LungCURE, the first standardized multimodal benchmark built from 1,000 real-world, clinician-labeled cases across more than 10 hospitals. We further propose LCAgent, a multi-agent framework that ensures guideline-compliant lung cancer clinical decision-making by suppressing cascading reasoning errors across the clinical pathway. Experiments reveal large differences across various large language models (LLMs) in their capabilities for complex medical reasoning, when given precise treatment requirements. We further verify that LCAgent, as a simple yet effective plugin, enhances the reasoning performance of LLMs in real-world medical scenarios.
Journal IF: unknown (could not be matched)Reference: Hao F, Yang J, Zhu Y, Yu Z, Wu Q, Wang Y, et al. LungCURE: Benchmarking Multimodal Real-World Clinical Reasoning for Precision Lung Cancer Diagnosis and Treatment. Proceedings of the 34th ACM International Conference on Multimedia. 2026 Nov 9 [Epub ahead of print]. doi:10.1145/3767308.3838653.Checked: Abstract only - Abstract
Imaging-defined high-risk phenotypes may capture biologically aggressive disease states, but translating them into therapeutic hypotheses remains challenging. We developed ParetoRx, a radiogenomic imaging-informatics framework integrating computed tomography radiomics, pathway-level transcriptomic analysis, multiobjective Pareto optimization, and model-free Q-learning for non-small cell lung cancer (NSCLC). Pretreatment CT data from 402 patients were used for model development and internal validation, with 110 patients forming a held-out external cohort. The final radiomics model achieved an external C-index of 0.707 (95% CI 0.627-0.783); high-risk patients had worse overall survival (HR 3.461, 95% CI 1.674-7.153), with minimal change after adjustment for histology (HR 3.444, 95% CI 1.662-7.137). Gene set enrichment and single-sample pathway analyses identified exploratory associations involving E2F targets, MYC targets V1, inflammatory response, and hypoxia, although no pathway met FDR < 0.05. LINCS L1000 perturbational profiles were evaluated using transcriptomic reversal, pathway specificity, and a feasibility-related proxy, with 4188 compounds retained in the common complete-case analysis. Multiobjective Pareto optimization was combined with three-split consensus ranking and sequential policy evaluation. TWS-119 showed consistent prioritization across all three splits (ranks 1, 3, and 4) and was Pareto-optimal in each split. In simulation-based evaluation, model-free Q-learning achieved higher terminal-goal attainment than myopic and static policies across all three splits, with improvements of approximately 6.2-13.3 percentage points over the myopic comparator. These findings support ParetoRx as an exploratory framework for biologically informed compound prioritization and computational sequential decision-making, while independent drug response and clinical validation remain necessary.
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: Kawahara D, Kishi M, Kadooka Y. ParetoRx: A Multiobjective Radiogenomic Framework Integrating Pareto Optimization and Q-Learning for Exploratory Compound Prioritization in Non-Small Cell Lung Cancer. J Digit Imaging Inform med. 2026 Nov 5 [Epub ahead of print]. doi:10.1007/s10278-026-02451-8.Checked: Abstract only