Accurate Diabetic Foot Ulcer Segmentation: A Human-Machine Collaborative Approach with EfficientNet and Self-ONN FPN

Diabetes mellitus is a chronic metabolic disease that affects millions of people worldwide and often leads to diabetic foot ulcers (DFUs). DFUs, which are a major source of morbidity and mortality and are brought on by neuropathy, ischemia, and poor wound healing, significantly raise the risk of lower limb amputations. Effective treatment of DFUs depends on their timely and accurate identification. However, the visual inspection-based clinical procedures used today are subjective and prone to mistakes. Using computer-aided approaches is a possible alternative. In this study, we introduced CFUD-3010, a new and extensive DFU segmentation dataset. 3,010 tagged photos were produced by merging two publicly accessible datasets, the Chronic Wound Dataset and DFU 2020. Because the DFU 2020 dataset lacked segmentation annotations, we used a joint human-machine method to create ground truth masks. In addition, we provide a new Self-Organized Operational Neural Network (Self-ONN)-based decoder and a pre-trained EfficientNetB3-based encoder for the DFU segmentation tasks. By improving heterogeneity and network variety while maintaining computational efficiency, self-ONNs get around the drawbacks of conventional convolution-based models. Using a STAPLE-based methodology, our model achieved precision of 87.918% and a Dice Similarity Coefficient (DSC) of 86.379%. The suggested model was tested on 200 more photos for external validation in order to assess its generalizability. It surpassed current standards with precision of 92.298% and a DSC of 91.217%. Our method demonstrates the ability of sophisticated deep learning models to deliver precise, automated DFU segmentation, which can significantly enhance clinical evaluations and patient outcomes.