Detection of diabetic foot ulcers using artificial intelligence

Diabetic foot ulcer (DFU) is a severe health complication in diabetic patients that significantly increases the risk of infection, limb loss, and mortality. If not diagnosed early, it substantially reduces patient quality of life and imposes a considerable burden on healthcare systems. In this study, deep learning-based classification models were evaluated for the accurate and early detection of DFU. DFU images, labeled by expert physicians and obtained from the Kaggle platform, were analyzed using four different convolutional neural network (CNN) architectures: RegNetY-400MF, RegNetY-800MF, EfficientNet-B0, and EfficientNet-B1. The training process was enhanced through transfer learning and data augmentation techniques, while image preprocessing steps such as contrast enhancement and noise reduction further improved classification performance. The analysis results demonstrated that EfficientNet-B1 and RegNetY-800MF outperformed the others, exhibiting high accuracy, low loss, and strong generalization capabilities. The EfficientNet-B1 model achieved the highest results with an accuracy exceeding 99%, whereas RegNetY-800MF also showed consistent performance on both training and testing datasets. Although the RegNetY-400MF model achieved high success during training, it exhibited fluctuations during testing. In contrast, the EfficientNet-B0 model, with its more compact architecture, was evaluated as a suitable alternative for systems with limited computational resources. The discussion section revealed that all models achieved sensitivity rates close to 100%, ensuring that no ulcerated cases were missed. In terms of ROC curves and AUC values, EfficientNet-B1 and RegNetY-800MF stood out as reliable structures potentially applicable in clinical settings. Moreover, the high specificity rates indicated a minimized false positive ratio. In conclusion, this thesis demonstrates that artificial intelligence-assisted solutions for DFU diagnosis are applicable in both academic research and clinical practice.