Computer-aided diagnostic (CAD) systems leveraging artificial intelligence (AI) hold significant promise for enhancing image processing capabilities in prostate cancer detection. This study aims to develop a CAD system using three-dimensional multiparametric transrectal prostate ultrasound (3D mpUS) to accurately localize clinically significant prostate cancer (csPCa). We included 252 patients diagnosed with csPCa (Gleason Group≥2) scheduled for radical prostatectomy (RP) and 83 “negative” patients who showed suspicion for prostate cancer but had negative MRI results (PIRADS≤2) or negative systematic biopsies. These patients underwent 3D mpUS, which combined 3D grayscale imaging, 4D contrast-enhanced ultrasound (CEUS), and 3D shear wave elastography (SWE) using an automated acquisition sequence. Histopathology from whole-mount RP specimens provided the ground truth for training the CAD system, which was registered to mpUS images using custom software. We extracted quantitative features from SWE, B-Mode, and CEUS to establish key imaging biomarkers for prostate cancer. These features were optimally combined through a dedicated AI model trained for voxel-based detection of csPCa, validated using a 5-fold cross-validation method applied to three random dataset splits. The CAD system achieved a sensitivity and specificity of 0.82 for csPCa detection, with an area under the receiver operating characteristic curve (AUROC) of 0.90. The negative predictive value was 0.65, and the positive predictive value was 0.92, although the dataset’s csPCa prevalence of 70% may not reflect the general population. Overall, the developed CAD system demonstrates promising voxel-level detection accuracy for csPCa, paving the way for future lesion-level validation and prospective clinical trials to evaluate its diagnostic performance and clinical applicability.
