Diabetic Foot Ulcer Classification Models Using Artificial Intelligence and Machine Learning Techniques: Systematic Review
Summary: This systematic review examines the performance and limitations of machine learning (ML) models developed to classify and prognosticate diabetic foot ulcer (DFU) outcomes. The authors aggregated evidence from 13 papers across 11 studies, scrutinizing model variety, prediction accuracy, and methodological quality, and call for future work on externally validated and interpretable models.
Key Highlights:
- Scope: 11 studies (13 publications) were included, assessing ML models for outcomes such as wound healing, lower extremity amputation (LEA), and mortality.
- Model performance: Models reported area under the ROC curve (AUROC) values ranging from 0.56 to 0.94; many reached ≥0.8, indicating good discrimination.
- Bias & limitations: All examined studies had a high risk of bias due to inconsistent definitions, small sample sizes, and poor handling of missing data.
- Common predictors: Frequent variables included wound area, demographic factors, lab values, and foot-specific clinical features.
- Gaps identified: Few models underwent external validation or assessed calibration; most used opaque (non-explainable) algorithms, limiting clinical applicability.
- Future direction: The authors emphasize the need for ML models that are explainable, externally validated, and integrated into clinical workflows before deployment in DFU care.
Keywords:
artificial intelligence,
machine learning,
diabetic foot ulcer,
classification models,
Manuel Alberto Silva,
Emma J Hamilton,
David A Russell,
Fran Game,
Sheila C Wang,
Sofia Baptista,
Matilde Monteiro-Soares