Comparative Image Quality of Deep Learning Reconstruction Against Conventional Reconstruction for T2-weighted Prostate Imaging: A Systematic Review and Meta-Analysis

Abstract

Context: Deep learning reconstruction is increasingly proposed to accelerate prostate T2-weighted magnetic resonance imaging (MRI) acquisition without loss of diagnostic quality. However, the magnitude of image quality benefits and the influence of deep learning network design remain unclear. Objectives: We aimed to compare the image quality of deep learning versus conventional prostate T2 MRI reconstruction and assess how acceleration factors and network architecture influence performance. Methods: A PRISMA search of PubMed/Medline, Embase, Web of Science, and Scopus (22 February 2026) identified studies comparing deep learning with conventional T2 MRI reconstruction. Paired standardized mean change (SMCC) values were pooled with a three-level random-effects model from extracted paired image-quality scores; meta-regressions examined scan-time acceleration. Small-study bias was tested with a precision-effect method. Results: Five studies (602 participants; 20 comparisons) included in the study. Overall deep learning reconstruction benefit was +0.14 standard deviation (SD) [95% confidence interval (CI) -2.00 to 2.28; I2 = 99 %]. At equal scan times deep learning reconstruction improved quality by +2.09 SD; for each fold increase in acceleration, image quality declined by 0.48 SD (95% CI -0.91 to -0.05; P = 0.030). C-SENSE AI at acceleration factors of 1.7, 3.4, and 4.8 demonstrated superior performance, whereas CycleGAN showed inferior results. Small-study effects were evident. Conclusion: Deep learning reconstruction can increase prostate T2-weighted MRI speed without significant changes in overall image quality; however, overly aggressive implementations may lead to image degradation. Careful validation and protocol-specific tuning are therefore essential prior to clinical adoption.