Development of a deep learning survival model to predict the individual risks of peritoneal metastasis of colorectal cancer

Background : Peritoneal metastasis (PM) has been considered to be the terminal stage of colorectal cancer (CRC) due to poor prognosis. We purposed to construct an AI model of clinicopathological parameters to predict the survival prognosis of PM in CRC. Methods : Our model was constructed through modifying a classic neural network with COX proportional hazards as loss function in the training cohort. It is able to predict the overall survival (OS) of each individual patient with clinicopathological parameters. Multivariate analysis was conducted to identify independent risk factors for the prognosis of PM patients. Results and Conclusion : In the testing cohort, the deep learning model show good performance with the c-index of 0.76 and brier score of 0.20. Conclusions : We have developed a deep learning model to predict the survival of individual patients precisely. It can provide evidence to apply personalized treatments and assisted surgeon to select optical treatments for CRC patients with PM.