A team led by the Second People’s Hospital of Lianyungang conducts a systematic review and meta-analysis assessing machine learning algorithms applied to multiparametric MRI for prostate cancer diagnosis, pooling sensitivity, specificity, and AUC across twelve studies to quantify accuracy in differentiating benign versus malignant lesions and identifying clinically significant tumors.

Key points

• Pooled sensitivity of 0.92 and specificity of 0.90 for benign versus malignant detection, with AUC of 0.96 across five studies.
• Machine learning models integrate features from T2-weighted, diffusion-weighted (ADC), and dynamic contrast-enhanced MRI sequences to assess lesion heterogeneity.
• Seven studies focused on Gleason score ≥7 csPCa, yielding pooled sensitivity 0.83, specificity 0.73, and AUC of 0.86.

Why it matters: These findings demonstrate that AI-enhanced MRI can outperform conventional PI-RADS, paving the way for more accurate, noninvasive prostate cancer screening.