Enhanced Diabetic Foot Ulcer Detection Using Deep Learning: The Impact of Efficientnet on Improving Classification Precision

This study presents a deep learning-based model for the automated detection of DFUs using foot images. For the purpose of model training and evaluation, a dataset of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1, 0 5 5}$</tex> images including 512 images of abornal footer with ulcers and 543 images of normal and healthy page of 512 were used. To achieve the 97 % accuracy, the EfficientNetB3 architecture was chosen because it is efficient and capable of inducing transfer learning. It showed very good performance in discriminating between healthy and abnormal feet with precision and recall scores greater than <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 5}$</tex> for both classes. Data augmentation and fine tuning was used to aid generalization with little overfitting. Resultantly, the prediction was evaluated using a confusion matrix and classification report with only a few misclassifications which confirmed the model's reliability. The results show that this model is appropriate as the model for real world DFU detection and can be applied to real world clinical setting or mobile health application. Future work may include increasing the size and diversity of the datasets used to enhance the model and deploying it as a real time application in healthcare.