AI and Cancer Research ── Lung ── 2026-10-25
Lung cancer: treatment, trials and AI
- Abstract
Background: Pulmonary neuroendocrine carcinomas (pNECs) are a rare and heterogeneous group of malignancies, ranging from low-grade carcinoids to high-grade small-cell lung carcinoma (SCLC) and large-cell neuroendocrine carcinoma (LCNEC). Prognostic tools for these tumors are scarce and have rarely been validated in external cohorts.
Objectives: This study aimed to develop a Transformer-based prognostic prediction model for pNEC and validate it in an independent external cohort, to improve survival estimation and support individualized treatment decisions.
Methods: This retrospective two-center cohort study included 823 patients with surgically resected pNEC (2001-2023). Models were developed in Center 1 (n=374; randomly split 8:2 into training and internal test sets) and externally validated in an independent Center 2 cohort (n=449). Three algorithms—logistic regression (LR), decision tree (DT), and the Transformer-based tabular foundation model TabPFN—were compared for estimating overall survival (OS) and disease-free survival (DFS) at 1-, 3-, and 5-year landmarks, with evaluation by receiver operating characteristic, calibration, and decision curve analyses, and SHapley Additive exPlanations (SHAP) analysis for interpretability.
Results: TabPFN achieved the highest area under the curve (AUC) in most tasks in both test sets, with task-dependent exceptions: in the internal test set, logistic regression performed best for 1-year OS (AUC 0.885), while TabPFN performed best for the remaining tasks, including 3-year OS (0.880), overall DFS (0.810), and 1-year DFS (0.892); in the external test set, TabPFN remained best for 1- and 3-year OS and for overall, 1-, and 3-year DFS (0.747-0.802), whereas logistic regression performed best for 5-year OS and DFS. External calibration was less satisfactory. Disease stage, pathological subtype, and age were the key prognostic predictors, and TabPFN was more accurate for low-grade subtypes. SHAP analysis quantified the contribution of each feature and made the model interpretable.
Conclusions: The Transformer-based model improves survival estimation in pNEC, with task-dependent advantages over conventional methods and confirmed transportability to an independent institutional cohort. Prospective multicenter validation and calibration refinement are needed before broader clinical use.
Journal IF-equivalent: 2.7 (OpenAlex 2-year mean citedness, value as of 2026-10-02, retrieved 2026-10-03; not the official Clarivate IF)Reference: Liu L, Liang L, Wang D, Zhang J, Zhao K, Guo C, et al. Prognostic prediction of pulmonary neuroendocrine carcinoma using a transformer-based artificial intelligence model: a two-center retrospective cohort study. Front Oncol. 2026 Oct 23;16:1948262. doi:10.3389/fonc.2026.1948262.Checked: Abstract only - Abstract
Background: Bone metastasis is a common and clinically important complication of non-small cell lung cancer (NSCLC) and is associated with poor survival. Although positron emission tomography-computed tomography (PET-CT) provides comprehensive skeletal assessment, its cost and radiation exposure may limit routine use. Therefore, an accessible approach based on routinely available clinical and chest CT-related information may help identify patients at higher risk who could benefit from further skeletal imaging. This study aimed to develop an effective predictive model for bone metastasis in patients with NSCLC.
Methods: This study retrospectively analyzed data from 17,646 Asian NSCLC patients in the Surveillance, Epidemiology, and End Results (SEER) database (2000–2021) as a training set and used an independent external cohort of 308 NSCLC patients from Shanghai Chest Hospital (2022–2023) for validation. Univariate and multivariate logistic regression (LR) analyses and Collinearity analysis were performed for variable selection. A novel DualDrop-Gated Transformer (DGT) model was developed and compared with LR, extreme gradient boosting (XGBoost), LightGBM, TabTransformer, and FTTransformer. Performance was assessed by receiver operating characteristic (ROC), calibration, and decision curves.
Results: Age, sex, T stage, N stage, histological subtype, and lung metastasis status were identified as significant predicting variables. The DGT model achieved an area under the curve (AUC) of 0.947 in the training set, 0.936 in the internal test set, and 0.879 in the external validation cohort with balanced sensitivity (0.856) and specificity (0.934), significantly outperforming other models (P<0.01). Ablation experiments demonstrated the significant contributions of gating mechanisms and dropout to model performance. A nomogram based on the DGT model was constructed, with reasonable calibration, and decision curve analysis (DCA) revealed good clinical utility.
Conclusions: A novel DGT prediction model for bone metastasis in patients with NSCLC based on a highly generalizable and interpretable deep learning framework was developed and validated using 6 predictors, which may have potential application value in early screening, decision‐making, and individualized management for lung cancer patients.
Journal IF-equivalent: 1.9 (OpenAlex 2-year mean citedness, value as of 2026-10-12, retrieved 2026-10-13; not the official Clarivate IF)Reference: Wang X, Yu H, Shen Y, Qian Q, Gao X, Xie Z, et al. A DualDrop-Gated Transformer model for predicting bone metastasis in Asian patients with non-small cell lung cancer: a real-world study with external validation. Quant Imaging Med Surg. 2026 Nov;16(11):863-3. doi:10.21037/qims-2026-0574.Checked: Abstract only - Abstract
Background: Accurate prediction of local recurrence (LR) after stereotactic body radiation therapy (SBRT) remains challenging in patients with lung cancer. The aim of the present study was to develop an artificial neural network (ANN)-based prediction model for predicting 3-year LR following SBRT. Materials and methods A total of 127 lung tumors treated with SBRT were retrospectively analyzed. Predictors selected by multivariable logistic regression using the Akaike information criterion were used to construct Model 1. Model 2 was developed by additionally incorporating the maximum biologically effective dose to the planning target volume (BEDPTV max ) and histology into Model 1. Both models were developed using an ANN with internal bootstrap validation. Model discrimination, calibration, and clinical utility were assessed using receiver operating characteristic analysis, calibration plots, and decision curve analysis. The predictive performance of the ANN models was compared with that of a conventional logistic regression and three radiobiological models using the same dataset.
Results: Model 2 demonstrated significantly improved discrimination than Model 1, with AUCs of 0.892 (95% CI, 0.838-0.944) and 0.791 (95% CI, 0.712-0.856), respectively ( P < 0.05). Both ANN models showed a general agreement between predicted and observed probabilities and provided net benefit across clinically relevant threshold probabilities. Model 2 also demonstrated higher AUC than the conventional logistic regression and three radiobiological models.
Conclusion: ANN-based models have potential to predict 3-year LR in patients with lung cancer treated with SBRT and outperformed the conventional logistic regression and radiobiological models. External validation in independent cohorts is warranted.
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: Huang BT, Wang SY, Lin PX, Luo LM, Wang Y, Chen JZ, et al. Application of an interpretable artificial neural network for 3-year local recurrence prediction in lung cancer patients treated with stereotactic body radiation therapy. Heliyon. 2026 Nov;12(16):e45582. doi:10.1016/j.heliyon.2026.e45582.Checked: Abstract only - Abstract
Background: How treatment recommendations generated by large language models (LLMs) compare with real-world decisions in oncology remains poorly characterized, particularly in resource-limited settings.
Methods: In this single-center feasibility study, four LLMs (ChatGPT, Claude, Gemini, and Grok, accessed via consumer interfaces in June 2026) generated treatment recommendations for 27 consecutive newly diagnosed lung cancer cases treated at a Romanian oncology unit (June 2024–May 2026). Each model’s output and the treatment actually delivered were formatted identically (135 recommendations total), presented in randomized order, and scored for appropriateness (1–5) by a reviewer blinded to the source, with three quality flags. A second independent reviewer re-scored all recommendations to assess inter-rater reliability. Scores were compared using the Friedman test.
Results: Mean appropriateness was highest for the real-world decision (4.52), followed by ChatGPT (4.48), Claude (4.44), Grok (4.41), and Gemini (4.37), with no significant difference (Friedman χ24 = 1.99, p = 0.74). No recommendation scored below 3. Inter-rater agreement was high in raw terms (93.3% within one point) but only fair after chance correction (weighted κ = 0.26), reflecting a ceiling effect.
Conclusions: AI-generated recommendations were rated similarly to real-world decisions. The evaluation method proved feasible in a resource-limited setting; the small sample and ceiling effect preclude equivalence claims, and the findings are hypothesis-generating.
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: Popescu A, Jidveian-Popescu M, Mahler B, Olteanu R, Clem E. AI-Generated Treatment Recommendations in Newly Diagnosed Lung Cancer: A Single-Center Feasibility Study in a Resource-Limited Setting. Curr Oncol. 2026 Oct 21;33(10):640. doi:10.3390/curroncol33100640.Checked: Abstract only