Hanchang Wu, Fang Liu, Qingsong Yang, Haihu Chen, Yan Wang, Xiaoguang Yang, Peng Xia, Lei Fang, Gang Li, Jing Yang, Yindeng Luo, Jing Li, Xu Fang, Xuedong Yang, Hui Jiang, Jianying Liu, Yong Yi, Xiangfei Chai, Jianping Lu, Jian Wang, Xu Gao, Chengwei Chen, Chengwei Shao, Yun Bian
Prostate MRI enables detection of clinically significant prostate cancer (csPCa), yet variability in PI-RADS scoring limits reproducibility and throughput. Here, we report the development and validation of an automated MRI-based decision aid (ProAI) that estimates patient-level risk of csPCa from biparametric MRI and supports routine reporting. Training, internal validation, and external testing spanned 7849 examinations across six centres and two public datasets. On pooled external tests, the system achieved a patient-level AUC of 0.93 (95% CI, 0.91-0.95), comparable to PI-RADS while improving inter-case consistency. In a multi-reader, multi-case study involving nine clinicians, assistance increased accuracy from 0.80 to 0.86 and reduced reading time. Prospective implementation in 1978 consecutive examinations-maintained performance (AUC 0.92) and was associated with a 32% reduction in radiology workload. Performance generalised to the TCIA cohort (AUC 0.83). These findings indicate that an automated MRI-based decision aid can standardise reporting and enhance efficiency across prostate cancer care pathways. This study was registered at ClinicalTrials. Trial number: ChiCTR2400092863.