Predicting Lung Cancer Survival after Curative Surgery Using Deep Learning of Diffusion MRI

Abstract The survival of lung cancer patients is expected differently according to the stage at diagnosis. However, each individual patient experiences different survival results even in the same stage group. DWI and ADC are two of widely used prognostic indicators for the prediction of survival in cancer patients. This study aims to develop a deep learning model that predicts the overall survival of non-small cell lung cancer patients using diffusion MRI. The study adapted a VGG-16 network and investigated the model’s performance using different combinations of DWI with/without ADC images. The survival model using deep learning of both DWI and ADC accurately predict the possibility of survival in five years after surgical treatment of NSCLC (up to 92%). The accuracy of results produced by the deep learning model can be enhanced by inputting precedented, proven functional parameters of ADC including the original images of DWI in survival prediction.