MRI-derived radiomics for prostate cancer active surveillance: a systematic review.

In 30 seconds

This PRISMA 2020 systematic review assessed MRI-derived radiomic and artificial intelligence models for selecting men with prostate cancer for active surveillance, detecting clinically significant cancer at baseline, and predicting progression. Seventeen studies from 12 unique cohorts were included; after resolving overlap, 12 cohort-level estimates represented 2,703 participants. Nine publications addressed progression, with reported AUCs ranging from 0.70 to 0.95. However, progression definitions differed across five non-equivalent approaches, preventing quantitative synthesis. Risk of bias was high or unclear in 16 of 17 studies, and no study reported external validation or calibration for progression.

Key findings

• After resolving overlapping reports, 12 independent cohort-level estimates included 2,703 participants.
• Across nine publications addressing progression, AUCs ranged from 0.70 to 0.95, but five non-equivalent endpoint definitions precluded pooling.
• Risk of bias was high or unclear in 16 of 17 studies; median METRICS score was 32.3 % (range 18.9-52.3 %).
• No study reported external validation or calibration for progression.

Why it matters

Radiomics and AI models may help inform active-surveillance decisions, but the review shows that reported performance is difficult to compare when progression outcomes are defined inconsistently. The lack of external validation and calibration also leaves uncertainty about how well these models would perform beyond the cohorts in which they were developed. Consistent, clinically meaningful outcomes are needed to assess their potential reliably.

What to keep in mind

The evidence base comprised 17 studies from 12 unique cohorts, with substantial overlap between publications; risk of bias was high or unclear in most studies. Heterogeneous progression definitions prevented quantitative synthesis.