Abstract PO-029: Survival prediction of non-small cell lung cancer by deep learning model integrating clinical and positron emission tomography data

The Cox proportional hazards model (CPH), the standard method for survival analysis in cancer patients, is difficult to use maximal information of positron emission tomography (PET) images containing information that reflects underlying pathophysiology. In this retrospective study, we aimed to investigate a deep learning model integrating clinical and PET image data to improve survival prediction in non-small cell lung cancer (NSCLC) patients. We developed a bimodal learning model thorough DeepSurv-based model using 3D-CNN-based 3D-Resnet model for image data and deep neural network (DNN) model for clinical data. DeepSurv-based model combining clinical and PET image data showed the best performance among the four models, the c-index of the training and testing sets reaching 0.898 and 0.768, respectively, followed by DeepSurv-based model for single modalities (clinical (0.763/0.740) or PET (0.773/0.743)) and CPH (0.726/0.736). Deep learning-based survival prediction with combination of two modalities may improve prediction accuracy in NSCLC patients. Citation Format: Sae-Ryung Kang, Seungwon Oh, In-Jae Oh, Jung-Joon Min, Hee-Seung Bom, Hyung-Jeong Yang, Guee-Sang Lee, Soo-Hyung Kim, Min Soo Kim. Survival prediction of non-small cell lung cancer by deep learning model integrating clinical and positron emission tomography data [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-029.