HMS-CTN: A Hybrid Multi-scale Convolutional Transformer Network with Feature Fusion and Support Vector Machine (SVM) for Enhanced Diabetic Foot Ulcer Detection

Diabetic foot ulcers (DFUs) are a common complication of diabetes, posing significant challenges for early detection and timely intervention. This paper presents a novel deep learning model, HMS-CTN, specifically designed for automated DFU detection. The proposed model integrates a custom convolutional module, ConvTransNet, which captures intricate spatial hierarchies and contextual dependencies at multiple scales. To further enhance feature extraction, a Multi-Scale Separable Convolutional Module (MS-SCM) is incorporated, along with a transformer encoder, to improve long-range dependency modeling. The MS-SCM captures multi-scale features efficiently, enhancing the model’s ability to recognize both fine-grained and high-level patterns. The transformer encoder improves long-range dependency modeling and global context, refining feature extraction for more accurate and robust predictions, particularly in clinical applications. The model architecture incorporates MobileNet for efficient computation, ensuring a lightweight design without sacrificing performance. A unique feature fusion approach combines outputs from both the MobileNet and ConvTransNet branches, enhancing the model’s ability to capture diverse feature representations. Finally, Support Vector Machines (SVM) are utilized for classification, further improving accuracy and robustness in predicting DFUs. We preprocessed the input images using CLAHE (Contrast Limited Adaptive Histogram Equalization) to enhance image features, particularly in low-contrast areas, which is crucial for improving model performance on medical imaging datasets. Additionally, we optimized CLAHE parameters using the Artificial Bee Colony (ABC) optimization algorithm, which fine-tuned the contrast enhancement process to further improve the model’s accuracy and computational efficiency. We conducted an ablation study to evaluate the contribution of each model component and configuration to the overall performance. Our extensive experiments on benchmark datasets, including thermography and foot ulcer images, demonstrate that HMS-CTN outperforms existing methods, achieving an impressive accuracy of 99.52% on foot images and 94.78% on the thermography dataset. The results indicate that the proposed model offers significant improvements in both accuracy and computational efficiency, making it a promising tool for real-world clinical applications, where quick and reliable DFU diagnosis is essential.