Preventive Detection of Diabetic Foot Ulcers Based on Plantar Thermal Image Using Deep Learning EfficientNet Model

Abstract

Diabetic foot ulcers (DFUs) are a significant complication of diabetes, affecting more than 537 million adults worldwide and leading to increased morbidity and mortality. This study investigates the application of infrared thermography (IRT) combined with the deep learning
model EfficientNet for early detection of DFUs through plantar thermal image analysis. This study used a dataset from IEEE Data Port, consisting of thermal images from 132 diabetic subjects and 44 non-diabetic subjects. The dataset underwent preprocessing steps, including balancing, labeling, and resizing to 512x512 pixels. The EfficientNet model was trained to classify ulceration risk into 3 class classification (healthy feet, diabetic feet with low ulceration risk, and diabetic feet with high ulceration risk), the model achieved 94.87% accuracy, 93.75% precision, 92.31% recall, and 92.20% F1 score. These findings highlight the challenges in developing an effective diagnostic system for DFUs and suggest that future research should focus on larger and more diverse datasets, better preprocessing techniques, and data augmentation to improve model performance.