Effectiveness of ConvNeXt variants in diabetic feet diagnosis using plantar thermal images

Diabetic foot ulcers are one of the most significant complications experienced by patients with Diabetes Mellitus (DM), which can result in amputation and demise; therefore, it is essential to diagnose them instantly. In this regard, the efficiency of ConvNeXt variants was examined for automatically detecting diabetic feet from thermal images. The dataset includes plantar thermal images of diabetic and non-diabetic subjects. After preprocessing and data augmentation using conventional techniques or fancy Principal Component Analysis (PCA), thermal images were classified utilising Fully Connected (FC) layers, Support Vector Machine (SVM), and Logistic Regression (LR), using features extracted by ImageNet pre-trained ConvNeXts. Results demonstrated that the model trained with ConvNeXt XLarge and LR for feature extraction and classification and generated training images by fancy PCA performed best and had 100% accuracy. In a second study, the number of training and test images was changed with each other to investigate the capability of the presented models and fancy PCA in achieving valid outcomes while having a limited dataset. Developing Computer-aided Diagnosis (CAD) systems using ConvNeXts to identify diabetic and healthy feet on plantar thermal images achieved desirable results, and their effectiveness in detecting diabetic feet can help clinicians diagnose early and prevent dire consequences.