Channel Attention Based on ResNet-50 Model for Image Classification of DFUs Using CNN

Diabetic Foot Ulcer (DFU) is one of the leading causes with high imputation rates, foot deformities, and even death. It is an open wound or skin infection that occurs in diabetic patients and leads to the lower limb. Despite preventive measures, DFUs are still a problem for patients and the healthcare system. Due to the rapid increase in DFUs, effective preventive strategies need to be addressed. With respect to the traditional clinical approach, in the present era, Machine Learning (ML) methods such as Convolutional Neural Network (CNN) approaches an essential role in DFU classification and achieving promising results. In this work, a CNN-based ResNet-50 with Channel Attention (CA) Network model is proposed to classify the foot images (healthy and diabetic). CA extracts channel-wise features used in ResNet-50 that employ a 3-layer bottleneck architecture to enhance the model's accuracy. The DFU dataset considered in this work is collected from a Kaggle repository having 1048 images, where 80% of the dataset is used for training and 10% each for testing and validation, respectively. Further, data augmentation is performed on the original dataset to prevent overfitting of the model. In the result analysis, training and validation accuracy attained 93% and 90%, respectively, which shows better performance than other State-Of-The-Art (SOTA) methods.