S2434
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
proven to improve the local control rate in patients with locally advanced rectal cancer (LARC). However, patients exhibit significant heterogeneity in postoperative recurrence risk. By integrating the AI model-extracted features of whole-slide images (WSI) after nCRT with Disease-Free Survival(DFS), this study aims to assess the postoperative recurrence risk of these patients and assist subsequent treatment. Material/Methods: To improve the prognostic assessment accuracy of patients with LARC after nCRT and assist subsequent treatment, this study developed a comprehensive analysis model that integrates AI-extracted pathological image features with postoperative survival prediction and validation. The research team screened and collected 534 WSIs from LARC patients who met the inclusion criteria from the clinical database of our institution as the training cohort. All included patients were cases who underwent surgical resection after receiving nCRT, with a postoperative follow-up duration of ≥ 12 months. Subsequently, the AI-extracted pathological features from the WSIs of the training cohort were correlated and integrated with the patients' DFS data to construct a survival prediction model. This model was ultimately used to assess the risk of tumor recurrence in patients from the independent validation cohort.A total of 229 LARC patients from the same institution were included in the independent validation cohort. Their baseline clinical characteristics were balanced with those of the training cohort. Qualified WSIs and complete follow-up data were also provided to ensure the reliability of the validation results.Survival analysis was performed using the Kaplan-Meier method with log-rank test, and the concordance index (C-index) served as the primary evaluation metric. Hazard ratios (HRs) and their corresponding 95% confidence intervals (CIs) were estimated via the Cox proportional hazards model. Results:
undersampling, four ensemble Extreme Gradient Boosting (XGBoost)–Neural Network (NN) models were trained. Performance was evaluated using K-fold cross-validation, independent testing, and False Discovery Rate (FDR)–corrected Wilcoxon tests. Results: The integrated model combining clinical, radiomic, and DL features achieved the best Area Under the Receiver Operating Characteristic Curve (AUC) and test accuracy (AUC = 0.84; accuracy = 85.71%), significantly outperforming models using clinical features alone (AUC = 0.64; accuracy =77.55%) or clinical + radiomics features (AUC = 0.62; accuracy =77.51%). WHO performance status was the most important clinical predictor, while the top imaging predictors were primarily DL-derived, suggesting complementary prognostic information between handcrafted and learned representations. Model stability was confirmed across all cross-validation folds. Conclusion: A hybrid artificial intelligence (AI) model integrating clinical, radiomic, and deep learning features substantially improves OS prediction in patients with non-small cell lung cancer undergoing radiotherapy. This approach supports individualized, risk-based treatment planning and highlights the potential of multi-modal feature integration for precision radiotherapy in locally advanced non-small cell lung cancer. References: 1.Mayerhoefer ME, Materka A, Langs G, et al. Introduction to radiomics. J Nucl Med. 2020;61(4):488- 495. doi:10.2967/jnumed.118.222893. 2.Hosny A, Parmar C, Coroller TP, et al. Deep learning for lung cancer prognostication: A retrospective multi-cohort radiomics study. PLoS Med. 2018;15(11). doi:10.1371/journal.pmed.1002711. 3.Braghetto A, Marturano F, Paiusco M, Baiesi M, Bettinelli A. Radiomics and deep learning methods for the prediction of 2-year overall survival in LUNG1 dataset. Sci Rep. 2022;12(1). doi:10.1038/s41598-022-18085-z. Keywords: NSCLC, Radiomics, Deep learning
Poster Discussion 1967
Feasibility Study on Prognostic Risk of Rectal cancer: AI Model-Extracted pathological features After Neoadjuvant Chemoradiotherapy Ziwei Dong, Shenlun Chen, Zeyang Lyu, Menghan Zhang, Jianan Liu, Jiazhou Wang, Weigang Hu, Zhen Zhang Department of Radiation Oncology, Fudan University Shanghai Cancer Center, Shanghai, China
Purpose/Objective: Neoadjuvant chemoradiotherapy (nCRT) has been
Made with FlippingBook - Share PDF online