EARLY DETECTION OF DIABETES MELLITUS VIA FOOT THERMAL IMAGING USING CONVNEXT AND EFFICIENTNETB3 ARCHITECTURES: A COMPARATIVE STUDY

Diabetes Mellitus (DM) is a chronic metabolic disorder that can lead to severe complications such as diabetic footsyndrome if not diagnosed early. Traditional diagnostic methods, though reliable, are often invasive and impracticalfor large-scale screening or continuous monitoring. This study proposes a non-invasive framework for early DM detectionusing thermal foot imaging combined with deep learning models. Specifically, it compares the performance of twoadvanced Convolutional Neural Network (CNN) architectures—ConvNeXt and EfficientNetB3—both optimized usingdata augmentation techniques.A variety of augmentation strategies, including geometric and intensity-based transformations, were ap- plied to improvethe model’s generalization. ConvNeXt, with its transformer-inspired enhancements, demonstrated superior accuracy, whileEfficientNetB3 offered balanced performance with fewer compu- tational resources. Evaluation metrics such as accuracy,sensitivity, precision, and F1-score were used to assess model effectiveness. Additionally, Grad-CAM heatmaps wereemployed to improve interpretabil- ity. A web-based interface was also developed using Flask to support real-time imageclassification. The results confirm the feasibility of AI-powered thermal imaging for early and efficient diabetes screening.