Optimized Classification with CBGRU-CapNet on Thermographic Analysis to Detect Diabetic Foot Ulcers with Precise Diagnosis

Diabetic foot ulcers (DFUs) are a severe complication of diabetes mellitus, often leading to lower limb or foot amputation due to infections and a high recurrence rate. Recent research focuses on early detection of underlying conditions that cause skin and tissue damage before visible lesions appear by analysing infrared thermograms of the plantar aspect of both feet. Most existing studies rely on the publicly available INAOE dataset, which includes thermographic images of both healthy and diabetic subjects. However, advancements in the field have been supported by the newly released STANDUP dataset, which offers a more extensive and diverse collection of thermograms. This study utilizes a broader dataset to improve DFU classification. Features are extracted from preprocessed input data using both statistical methods and deep feature extraction techniques. A novel deep learning model, CBGRU-CapNet − combining convolutional layers, bidirectional gated recurrent units (BiGRU), and capsule networks − is employed for classification. To enhance accuracy and reduce computational complexity, the study introduces a fine-tuning approach using the Modified Directional Transport Metaheuristic Algorithm (mDTMA). Experimental results demonstrate that the proposed model achieves superior predictive performance and robustness, making it highly suitable for accurate and consistent DFU detection across multiple evaluation metrics. The results demonstrate that the proposed model provides superior predictive power and robustness, indicating its suitability for tasks requiring high accuracy besides consistency across multiple estimation metrics