Automatic prediction model of overall survival in prostate cancer patients with bone metastasis using deep neural networks

Abstract Objectives Bone is the most common site of metastasis in prostate cancer (PCa) patients and correlated with poor prognosis increasing economic burden. Few studies have analyzed prognostic prediction for metastatic PCa assistance neural networks. Methods Four convolutional network (CNN) models are developed evaluated to predict overall survival (OS) bone metastasis. All CNN first trained 64 samples 10 samples; two use only scan images both clinical parameters (CPs). The predictions best compared those by urology surgeons on 20 test samples. Results Our can OS AUC=0.8022 using AUC=0.8132 CPs Youden indexes 0.6263 0.7142, respectively, which 0.3077 0.3131 higher than that urologists’ average index, indicate exhibit significant advantages. Conclusions suitable CPs. show better performance terms accuracy stability surgeons.