Automated assessment of diabetic foot ulcers: Estimation of PWAT categories using radiomic features and machine learning

Abstract

The assessment of diabetic foot ulcers is essential for guiding therapeutic interventions and predicting outcomes such as wound healing or the need for surgical procedures. In this study, an automated pipeline for quantifying six categories of the Photographic Wound Assessment Tool (PWAT) was developed and validated using a dataset of 1,153 expert-annotated photographic images of diabetic foot ulcers (DFUs). All images were uniformly resized to 256 × 256 pixels, and ninety-three radiomic features were extracted via the PyRadiomics library. Support Vector Machine, Random Forest, and XGBoost classifiers were trained on 75% of the dataset, with the remaining 25% held out for validation. Class imbalance was addressed using Synthetic Minority Oversampling Technique (SMOTE), random oversampling, and undersampling strategies. The optimal performance was achieved with XGBoost combined with random oversampling, yielding an accuracy of 78% and a macro-average F1-score of 0.77. These results demonstrate the feasibility of a reproducible, quantitative, and fully automated system for standardizing the evaluation of diabetic foot ulcers.