Radiomics predicts liver-cancer microvascular invasion well, but not yet for the clinic

Design

Systematic review and meta-analysis of diagnostic accuracy; random-effects pooling

Population

26 studies of CT- and MRI-based radiomics models in hepatocellular carcinoma

Primary outcome

Preoperative prediction of microvascular invasion

Effect

Pooled area under the curve 0.85 (0.82–0.88); clinicoradiomics models best

Microvascular invasion in hepatocellular carcinoma predicts early recurrence after curative treatment, but cannot be assessed reliably before surgery. This meta-analysis of 26 studies evaluated CT- and MRI-based radiomics models for predicting it preoperatively.

Pooled diagnostic performance was good, with an area under the curve of 0.85, and models that combined radiomics with clinical variables did better and more consistently than radiomics alone. But heterogeneity across studies was considerable and external validation scarce.

The honest position is that this is a promising research tool, not a clinical one. The models are not standardised, and almost none have been validated prospectively across centres and scanners. Treat radiomics for microvascular invasion as a signal of where preoperative imaging may go, not as something to report on a live case yet.

• Pooled area under the curve was 0.85 (95% CI 0.82–0.88) across 26 CT- and MRI-based studies.
• Combining radiomics with clinical variables outperformed radiomics alone (MRI 0.87 vs 0.85; CT 0.87 vs 0.81).
• Heterogeneity was considerable and external, multicentre validation was lacking.
• Do not use radiomics microvascular-invasion models in routine reporting until standardised and prospectively validated.

Why it matters

It marks the gap between impressive retrospective accuracy and the standardisation needed before a model can touch a real report.

Don't overread it

High pooled accuracy with little external validation often falls when a model meets new scanners and populations — this is not yet clinic-ready.

The statistics, in plain English

An area under the curve of 0.85 means good discrimination between tumours with and without microvascular invasion. But a pooled figure hides wide between-study variation, and without external validation the number may not hold on your scanner and population.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.