Diabetic Foot Ulcer Detection and Classification using ConvMixer Model

Diabetic foot ulcer (DFUs) are a common and significant consequence of diabetes, frequently resulting in severe illness and amputations if not promptly identified and treated. The growing occurrence of DFUs is a substantial obstacle to healthcare globally. Precise and prompt categorization of DFUs is essential for efficient treatment and avoidance of consequences. This research analyzed ConvMixer model for comprehensive classification designed to accurately evaluate DFU. The DFU images are preprocessed using the watershed algorithm that helps in separating different regions of an image, such as the ulcer and surrounding tissues, to improve the neural network’s ability to extract high-quality information. The framework utilizes depthwise and pointwise convolutions to mix spatial and channel-wise information in an efficient manner to effectively extract pertinent information from various clinical DFU dataset. The proposed work features a multi-layered depthwise and pointwise convolutional architecture, created especially to manage the complex patterns and variances found in various diabetic foot ulcer stages and variants (DFUs). By integrating these convolutions and skip associations, the network facilitates hierarchical extraction of features and robust representation of features. Experimental results on a substantial and varied DFU dataset demonstrate that the ConvMixer model achieves an impressive accuracy rate of $\mathbf{9 7. 1 6 \%}$, with a precision, F1-score and a sensitivity of $\mathbf{9 7 \%}$, $\mathbf{9 7 \%}$ and 99.08% respectively.