Early Detection of Diabetic Foot Ulcers: Deep Learning Classification using Thermograms (AlexNet Algorithm)

This study aims to evaluate and improve the performance of AlexNet algorithm in classifying diabetic foot ulcer (DFU) through thermograms. Diabetes Mellitus (DM) is one of the chronic and common illness on the planet. The primary complication of diabetes is DFU. Application of deep learning neural network in diagnosing diabetics’ feet can help in the early detection of DFU to prevent severe stage which may lead to amputation. In this paper, AlexNet deep learning neural networks was trained with the plantar thermogram dataset to classify the DFU. In addition, this paper also aims to investigate the effect of greyscale images in the image classification of thermogram on DFU by comparing the accuracy of image classification using colour thermograms and greyscale thermograms. The dataset used consists of 1866 coloured images which 890 images of plantar thermogram with negative DFU and 976 images of foot thermogram with positive DFU. The images were retrieved from Kaggle community with free and full access. In the experiment, data augmentation was performed on the dataset, which the 1866 images were converted into greyscale. As a result, the dataset images were doubled. At the end of this experiment, the improved network achieves an accuracy of 97.8% in classifying the thermograms.