ESTRO 2025 - Abstract Book

S3812

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Conclusion: We developed an AI framework to automatically quantify lymphocyte infiltration in CRC. These findings demonstrate that AI-identified TIL density is a robust predictor of overall survival in CRC regardless of tumour stage and could serve as a potential biomarker for personalized treatment. Future work includes exploring the interaction between the TIL density and genomic signatures to enhance precise chemo-radiotherapy planning. A real-world CRC cohort is being collected for further validation.

Keywords: lymphocyte infiltration, colorectal cancer

References: [1] Tan, M. and Le, Q.V. EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. Preprint at https://doi.org/10.48550/arXiv.1905.11946 (2019). [2] Kather, J. N., et al. 100,000 Histological Images of Human Colorectal Cancer and Healthy Tissue. v0.1, Zenodo, 7 Apr. 2018, doi:10.5281/zenodo.1214456. [3] Wang, A., et al. YOLOv10: Real-Time End-to-End Object Detection. Preprint at https://doi.org/10.48550/arXiv.2405.14458 (2024). [4] Simard, M., et al. Immunocto: a massive immune cell database auto-generated for histopathology. Preprint at https://doi.org/10.48550/arXiv.2406.02618 (2024).

3217

Proffered Paper Doses to cardiac substructures show superior predictive power and robustness over whole-heart for hs cTnT elevation in NSCLC radiotherapy Xinru Chen 1,2 , Xiaodong Zhang 1,2 , Ting Xu 3 , Radhe Mohan 1,2 , Ruitao Lin 4 , Rachel C Maguire 3 , Yao Zhao 1 , Efstratios Koutroumpakis 5 , Nicolas L Palaskas 5 , Anita Deswal 5 , Ali Ajdari 6 , Joshua S Niedzielski 1,2 , Sanjay S Shete 4,2 , Laurence E Court 1,2 , Jinzhong Yang 1,2 , Zhongxing Liao 3 1 Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, USA. 2 Graduate School of Biomedical Sciences, The University of Texas MD Anderson Cancer Center UTHealth Houston, Houston, USA. 3 Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, USA. 4 Department of Biostatistics, The University of Texas MD Anderson Cancer Center, Houston, USA. 5 Department of Cardiology, The University of Texas MD Anderson Cancer Center, Houston, USA. 6 Department of Radiation Oncology, Mass General Research Institute, Boston, USA Purpose/Objective: Cardiotoxicity is a critical concern for patients undergoing thoracic radiotherapy. This study compares the predictive power of whole-heart (WH) and cardiac substructure-based models for high-sensitive cardiac troponin T (hs-cTnT) elevation, 1 a biomarker for early detection of cardiac adverse events. Material/Methods: A retrospective cohort of 160 Non-Small Cell Lung Cancer (NSCLC) patients from a completed prospective trial (NCT00915005) 2 and a prospective cohort of 54 NSCLC patients enrolled in an ongoing trial (NCT05010109) were analyzed. The endpoint was hs-cTnT elevation (increase≥5ng/L) incidence per our prior report. 1 An in-house auto segmentation model delineated 19 cardiac substructures, including the WH, chambers, great vessels, valves, and coronary arteries, from each patient’s planning CT. 3 Dose-volume histogram (DVH) metrics, radiomic features, and dosiomic features were extracted from clinical plans, planning CTs, and clinical dose clouds, respectively. A 100 iteration Monte Carlo cross-validation (75%/25% split) was conducted within the retrospective cohort to mitigate random split bias. Logistic regression models using a different combination of inputs were compared to assess the predictive power between WH and cardiac substructures. A model using clinical factors only served as the baseline. Important features in prediction were identified via permutation-importance during training, and ensembled