Publication

Abstract : Diabetic Foot Ulcers (DFUs) is a consequence of diabetes, requiring early and accurate classification to prevent amputations. Non-ischemic wounds are easier to cure than ischemic. To plan treatment accordingly, manual differentiation is time-consuming and subject to variability, necessitating efficient classification methods. This research evaluates the performance of EfficientNet models, known for their scalability and efficiency, for differentiation of ischemic and non-ischemic DFU using a large dataset DFUC2021 of 15,683 images. We explore the family of EfficientNet variants, ranging from B0 to B6, to strike a balance between model complexity and computational efficiency. Utilizing transfer learning and augmentation, we analyze the models' performance in terms of metrics that include F1-score and recall. The B1 and B6 models achieved validation set classification accuracies of 93.01% and 93.06%, respectively, with highperformance metrics, indicating their superior ability in DFU Classification. We also assess computational requirements and model complexity, with B6 showing higher demands but yielding better accuracy. However, B1 emerged as the top performer, being a lighter version with lower loss and almost equal validation accuracy compared to B6. This study provides valuable insights for researchers and clinicians in selecting the appropriate EfficientNet model for developing reliable clinical DFU decision support systems, considering accuracy and computational resources.