Could image-based artificial intelligence improve earlier recognition of diabetic foot ulcers (DFUs)? A systematic review and meta-analysis published in Diabetes Research and Clinical Practice found that deep learning (DL) models demonstrated high diagnostic accuracy for DFU detection, although performance was lower in multicenter datasets.

The analysis searched PubMed, Cochrane Library, Embase, and Web of Science through October 2025 for original studies evaluating image-based DL systems for DFU detection. Risk of bias was assessed using Quality Assessment of Diagnostic Accuracy Studies-Artificial Intelligence (QUADAS-AI). Diagnostic 2×2 tables were constructed, and pooled estimates were generated using a bivariate mixed-effects model.

Overall, 55 studies met the inclusion criteria. Of these, 32 studies contributed 87 diagnostic 2×2 validation tables for quantitative synthesis. Across overall validation datasets, pooled sensitivity was 0.96 (95% confidence interval [CI], 0.94-0.98) and pooled specificity was 0.97 (95% CI, 0.94-0.98), indicating strong discrimination between ulcerated and non-ulcerated images.

Performance was lower in multicenter datasets, where pooled sensitivity was 0.87 (95% CI, 0.68-0.96) and pooled specificity was 0.92 (95% CI, 0.82-0.97). These findings suggest that external generalizability may remain a key challenge when models are applied across broader clinical settings.

The analysis concluded that image-based DL may offer a promising adjunct for DFU detection, but additional validation using individual-level data from geographically diverse populations is needed before broader clinical implementation.