JMIR Publications

ABSTRACT

Background:

Vascular compromise remains the leading cause of free-flap failure after microsurgical reconstruction. Artificial intelligence (AI)–based monitoring and prediction tools are emerging adjuncts for assessing viability and anticipating complications.

Objective:

This systematic review and meta-analysis quantify their diagnostic accuracy for flap surveillance and evaluates their role in forecasting adverse outcomes.

Methods:

Following PRISMA-DTA, PubMed, Embase and Cochrane were searched to August 2025 for studies developing AI tools for flap status assessment or prediction. Pooled sensitivity, specificity, AUC and diagnostic odds ratio were estimated using a hierarchical random-effects model. Risk of bias and applicability were assessed with QUADAS-2. Prespecified subgroup analyses by modality, model-fit diagnostics, sensitivity analyses, and publication-bias testing with Deeks’ funnel plot were undertaken.

Results:

Of 512 studies screened, 12 met the inclusion criteria. Pooled analysis demonstrated an overall sensitivity of 0.75 (95% CI: 0.58–0.87), specificity of 0.88 (95% CI: 0.77–0.94), AUC of 0.89 (95% CI: 0.86–0.92), and DOR 22 (95% CI: 6-85). Image-based AI models demonstrated notably higher diagnostic accuracy, with a sensitivity of 0.93 (95% CI: 0.82–0.98), specificity of 0.93 (95% CI: 0.82–0.98), and AUC of 0.98 (95% CI: 0.96-0.99). In contrast, clinical-variable model yielded more modestly overall. For vascular compromise prediction, pooled sensitivity was 0.90 (95% CI: 0.79–0.95) and specificity 0.90 (95% CI: 0.78–0.95). Between-study heterogeneity was moderate, and QUADAS-2 indicated an overall moderate risk of bias.

Conclusions:

AI is a promising adjunct for flap surveillance and complication prediction, but its clinical utility and safe integration require confirmation in large, prospective, multicenter studies. Clinical Trial: N/A