Diabetic Foot Ulcer Detection on Mobile Platforms through Thermal Imaging and Deep Learning

Diabetes mellitus, causing substantial global mortality with millions affected in 2019, necessitates early detection methods to aid Diabetic Foot Ulcer (DFU) patients. We proposed a mobile application for early DFU identification using a Convolutional Neural Network (CNN) classifier. The application embedded a fine-tuned MobileNetV2 model specifically adapted for DFUs, trained on a plantar thermogram dataset through transfer learning. We evaluate the model’s classification performance pre- and post-embedded into the application. Results reveal that the pre-embedded model achieved 90% accuracy, 100% specificity, and 80% sensitivity. After embedding, the model retained 85% accuracy, 80% specificity, and 90% sensitivity. Notably, a slight accuracy decrease post-embedded is likely attributed to the normalization of the input image. This study emphasizes that the interplay between model performance and mobile embedded might influence the model’s performance. It underscores the meticulous consideration of mobile platform-specific processes crucial for accurate and effective early DFU detection.