The Value of a Clinical-Radiomics-Deep Learning-Pathomics Fused Models for Biochemical Recurrence Prediction After Radical Prostatectomy

Motivation: There are currently a few studies that utilize artificial intelligence technology to analyze both for precise prediction of prostate cancer prognosis. Goal(s): To explore the value of a multi-modality fused model integrating clinical feature, radiomics model, MRI DL model and pathomics model in enhancing predictive efficacy of post-RP BCR. Approach: The CRDH fused model was established by combining clinical model, radiomics model, MRI DL model and pathomics model with COX regression. Results: In the testing set, the CRDH model achieved a C-index of 0.87, significantly higher than pathological T stage, radiomics model, MRI DL model, pathomics model and clinical score (P<0.05). Impact: The multi-modality fused model incorporating clinical variable, radiomics model, MRI DL model, pathomics model was better than all single-modality models. Our model could assess the prognosis of patients with PCa after surgery, providing strong support for formulating subsequent treatment plans.