Diabetes Mellitus Classification Using CNN-Based Plantar Thermogram Analysis

Diabetes Mellitus (DM) is a chronic metabolic disorder that often causes serious complications, including neuropathy and lower extremity disorders, which impact the quality of life of patients. Early detection of DM is a major challenge due to limited data and the complexity of image analysis. This study proposes a plantar thermogram image-based approach to support non-invasive diagnosis of DM through the development of a Convolutional Neural Network (CNN)-based model and machine learning techniques. This model integrates data augmentation techniques, such as rotation, flip, and zoom, to improve image variation and model robustness. Two CNN architectures, InceptionV3 and ResNet-50, are used in the training process, followed by feature selection using the Chi-Square method and classification using the Random Forest algorithm. The results showed that the proposed model achieved the best performance with accuracy, F1-score, precision, recall, and AUC (Area Under Curve) of 99.6% each. This approach makes a significant contribution by showing improvement compared to previous methods, while opening up opportunities for the development of more efficient clinical applications in early detection and monitoring of DM.