Artificial intelligence for the prediction of synchronous and metachronous liver metastasis in colorectal cancer patients: a systematic review and meta-analysis.
Yassin Rahnama, Homayoun Pishraft-Sabet, Sara Eghbali et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis assessed imaging-based artificial intelligence models for predicting synchronous and metachronous liver metastases in patients with colorectal cancer. Across 21 included studies, pooled sensitivity was 0.80 (95% CI, 0.74-0.84) and pooled specificity was 0.81 (95% CI, 0.72-0.88); the summary receiver operating characteristic AUC was 0.84. The authors judged discrimination promising, but found methodological and reporting shortcomings, including a mean Radiomics Quality Score of 16.4 out of 36.
Key findings
- Across 21 studies, pooled sensitivity was 0.80 (95% CI, 0.74-0.84) and pooled specificity was 0.81 (95% CI, 0.72-0.88).
- The summary receiver operating characteristic AUC was 0.84 (p-AUC = 0.81), indicating reasonably high discriminative ability.
- At a pre-test probability of 38%, pooled PPV was 0.72 and NPV was 0.87.
- Mean Radiomics Quality Score was 16.4 out of 36, indicating poor methodological quality.
Why it matters
Imaging-based AI may help identify colorectal cancer patients at risk of synchronous or later liver metastases, an outcome relevant to assessment and follow-up. The pooled diagnostic performance is encouraging, but the reported quality concerns mean these estimates should be interpreted cautiously and do not establish how well models will perform in routine clinical settings.
What to keep in mind
The included studies had poor methodological quality overall, with frequent high or unclear risk of bias in the reference-standard and index-test domains.