Study of deep-learning models for detection and localization of early-stage clinically significant prostate cancer on mpMRI

The purpose of this study was to explore the best sequences for fully automated detection and localization of early-stage csPCa with a PSA range of 4-10 ng/mL.1628 prostate mpMRI examined with seven scanners were retrospectively enrolled. A training dataset (n=1428) was used to train models, i.e., a diffusion model was trained with diffusion-weighted imaging, and a biparametric model was trained with diffusion-weighted imaging and T2-weighted imaging. A hold-out test dataset (n=200) was reserved for validation. Our research shows that the diffusion model trained using ADC and DWI achieved best performance in detection and localization of early-stage csPCa.