Fuzzy Logic and Attention Gate for Improved U-Net with Genetic Algorithm for DFU Image Segmentation

Diabetic Foot Ulcers (DFU) require effective management through early diagnosis and appropriate treatment. Image segmentation is very important in the management of DFU to identify the ulcer area. The popularity of U -Net architecture in the field of medical image segmentation still has several shortcomings in managing extensive contextual information and the existence of activation functions that are susceptible to saturation. The contribution of this research is to modify U-Net by applying fuzzy activation functions, attention gates, and hyperparameter optimization using genetic algorithms to improve DFU image segmentation performance. Testing on public datasets shows significant improvements in evaluation metrics such as dice coefficient, IoU, precision, recall, and 11 score. These results show that the research carried out can improve the diagnosis and treatment of diabetes patients through more accurate and stable medical image segmentation. Evaluation metrics show that after optimization with a genetic algorithm, the model achieves significant performance improvements compared to previous results. The test results also show that the model tested on the DFU2022 dataset has higher performance in terms of dice coefficient, IoU, precision and recall. Future research could consider using different data sets and additional approaches, as well as further evaluation to improve model performance.