Clinical Decision Support System for Diabetic Foot Ulcer Detection and Classification Model Using Fine-Grained Deep Transfer Learning Model

Diabetic Foot Ulcers (DFU) remain a significant health issue for diabetes patients, necessitating early and accurate detection for rapid intervention. Because of insufficient blood circulation, the reappearance of these ulcers can result in 84% of lower limb amputation and even lead to mortality. Severe diabetes risk patients need costly medications, routine check-ups, and appropriate personal cleanliness to avoid DFUs that affect 15–25% of diabetics. Early detection, accurate diagnosis, proper care, and rapid response can eliminate amputations and death rates by early and robust DFU detection from image analysis. Therefore, efficient DRC detection is a necessity for enhanced patient care. Recently, Deep Learning (DL) systems have attained substantial consideration in the prognosis and diagnosis of DFU detection through different medical imaging conditions. This study develops an Automated Diabetic Foot Ulcer Detection and Classification utilizing Snow Ablation Optimization with Deep Learning (ADFUDC-SAODL) approach. The ADFUDC-SAODL approach enables early detection and classification of DFU leveraging a fine-grained DL approach. In the ADFUDC-SAODL approach, the MResCaps-based feature extraction approach is employed. The MResCaps method is an enhancement of the Capsule Network (CapsNet), which comprises a convolutional layer, a Primary Capsule (PC) layer, and a Digit Capsule (DigitCaps) layer. As the CapsNet model makes use of a single convolution layer, fundamental features of the image are extracted. To extract detailed features and enhance the classification efficiency of the CapsNet method, residual blocks are utilized rather than the convolution layer. Besides, a Convolutional Recurrent Neural Network (CRNN) approach was performed for the classification of DFU. At last, the Snow Ablation Optimization (SAO) method adjusts the hyperparameter values of the CRNN method optimally and leads to enhanced classification performance. An extensive set of simulations was involved to exhibit the encouraging outcomes of the ADFUDC-SAODL methodology. The simulated outcomes implied the proficient performance of the ADFUDC-SAODL methodology across other recent models.