Evaluating deep learning and radiologist performance in volumetric prostate cancer analysis with biparametric MRI and histopathologically mapped slides

This study focuses on leveraging artificial intelligence (AI) to enhance the predictive accuracy of bpMRI, aiming to align its performance more closely with that of WMH-based tumor volume estimations.By comparing AI-enhanced bpMRI interpretations with traditional radiologist contours, this research seeks to underscore potential improvements in treatment planning and patient outcomes. Materials and methods Study sampleThis HIPAA-compliant retrospective study was conducted with the Institutional Review Board approval.Each patient included in the study provided a written informed consent and were enrolled in at least one of the following protocols: NCT00102544, NCT02594202, NCT03354416.All patients who received mpMRI scan, and subsequently had radical prostatectomy (RP) at our institution between January 2014 to November 2021 were considered for inclusion.Patients who had received other PCa treatments before the MRI, those included in the AI training dataset, those without intraprostatic lesions on MRI, those lacking prospective lesion segmentations, and those without detailed histopathological mapping of the whole-mount specimens were excluded from the study.The final study sample consisted of treatment-nave patients with MRI-pathology registered data (Fig. 1). mpMRI acquisition and interpretationAll patients underwent imaging on a 3T MRI (Achieva 3.0T TX scanner, Philips, Best, The Netherlands or Ingenia Elition 3.0T X, Philips, Best, The Netherlands or Verio