Jeong-Eun Moon, Yong-Jin Cho, Se-Yeol Rhyou, Sang-Hoon Hong, Journal of Internet Computing and Services, Vol. 26, No. 2, pp. 167-178, Apr. 2025
10.7472/jksii.2025.26.2.167, Full Text:  HTML
Keywords: Diabetic Foot Classification, AI-Powered Healthcare, Object Detection, Convolutional Neural Network (CNN), explainable artificial intelligence (XAI)

Abstract

The rising prevalence of diabetes and its associated complications necessitates innovative solutions for the classification and management of diabetic foot conditions. This study presents DiabetesNet, an AI model specifically designed for accurate and efficient diabetic foot classification, optimized for real-world deployment. YOLOv5n is employed for precise foot detection, while EfficientNet-B0 ensures lightweight yet robust classification. To address background-induced bias, the detected foot regions are cropped prior to classification, focusing the model solely on relevant features. Explainable AI (XAI) tools, such as Grad-CAM, enhance interpretability by enabling healthcare professionals to understand the model’s decision-making process. Experimental evaluations demonstrated exceptional performance, achieving a mean Average Precision (mAP) of 0.9949 for single-class detection and 0.9905 for dual-class detection. Sensitivity and specificity metrics were similarly high, showcasing the model’s robustness and reliability. Using data from 200 participants, this study validates the feasibility of the proposed approach. While these results are promising, further research with larger and more diverse datasets is required to confirm the system’s generalizability in clinical settings. DiabetesNet is highlighted as a lightweight and effective tool for diabetic foot classification, with significant potential to improve early detection and intervention in diabetic foot care.

Statistics
Cite this article