Prostate Cancer Risk Stratification Using Artificial Intelligence-Based Multimodality Integration: A Retrospective Two-Center Study in 1,442 Patients

Background: Pretreatment assessment of prostate cancer (PCa) aggressiveness is important for determining appropriate treatment option and prognostic prediction. Therefore, we employed an artificial intelligence (AI)-based multimodality integration to obtain a precisely informed risk stratification tool (PI-Risk) for PCa aggressiveness.   Methods: This study included total 1,442 biopsy-naive patients from two medical institutions, consisting of 671 datasets for model training and an independent set of 232 patients for internal test and 539 patients for external validation. The PI-Risk is designed to discriminate low-risk (Gleason score 6 and biopsy-benign), intermediate-risk (Gleason score 7) and high-risk (Gleason score ≥ 8) PCa by integrating large-scale multimodal information from clinical factors, multiparametric MR radiomics and high-dimensional deep learning-transformed imaging features extracted at intratumorous and peritumorous regions. Algorithm analysis and modelling was performed using an auto AI framework that offers us to deploy algorithm stacking and multimodality-integration. Findings: In 232 internal-tested data, PI-Risk achieved excellent discriminability (area under the receiver operating characteristic curve [AUC] > 0.9) both in low-risk vs intermediate-risk and low-risk vs high-risk disease; and moderate discriminability (AUC > 0.7) in intermediate-risk vs high-risk disease. In 539 external-validated data, the performance of PI-Risk retained excellent (AUC > 0.9) in low-risk vs high-risk disease, and retained good (AUC > 0.8) in low-risk vs intermediate-risk disease; and was moderate (AUC > 0.7) between intermediate-risk vs high-risk disease. In follow-up, PCa patients with different PI-Risk score showed significantly different recurrence rate after radical prostatectomy. Interpretation: PI-Risk offers a noninvasive alternative tool to stratify PCa aggressiveness. Funding Statement: 1. Contract grant sponsor: Key research and development program of Jiangsu Province; contract grant number: BE2017756 (to Y.D.Z.) 2. Contract grant sponsor: Suzhou Science and Technology Bureau-Science and Technology Demonstration Project; contract grant number: SS201808 (to X.M.W) 3. Contract grant sponsor: National Key RD contract grant number: 2017YFC0114300 (H.C.H.) Declaration of Interests: The authors declare no competing interests. Ethics Approval Statement: Ethics committee approval was granted by local institutional ethics review board with a waiver of written informed consent. All procedures performed in studies involving human participants were in accordance with the 1964 Helsinki declaration and its later amendments.