Multi-class classification of diabetic foot ulcer using faster R-CNN based on wagner grading scale

Abstract

Diabetic Foot Ulcers (DFU) are a major complication of Diabetes Mellitus (DM), that affects the lower extremities. It has been estimated that patients with diabetes have a lifetime risk of 15% to 25% in developing foot ulcers contributing to up to 85% of lower limb amputation due to failure to recognize and treat the foot ulcers properly. An extensive review of computerized techniques for the recognition of diabetic foot ulcers has been performed to associate the work done so far in this field. While performing the studies, it became clear that computerized analysis of diabetic foot ulcers is a relatively emerging field which is why related literature and research works are limited. There is also a lack of a standardized public database of diabetic foot ulcers and other wound-related pathologies. This dissertation report aims to develop an automated deep learning-based method for localizing and classifying foot ulcers of diabetic patients in different grades/scales. The classification of multi-class foot ulcers is based on the Wagner diabetic foot ulcer grade system using the Faster Region-based Convolutional Neural Network (Faster R-CNN) algorithm. The proposed automated classification process helps to take prompt therapeutic strategy and subsequent clinical treatment. In this approach, images of diabetic foot ulcers (DFU) are collected from a Primary Care Hospital in Kolkata. The annotation of DFU images is performed by the domain expert for localization and the annotated images are further classified for prediction. Faster R-CNN-based two architectures namely, resnet-101 and mobilenet-v3 network are employed for the prediction. The training of the deep learning model is validated using 5-fold cross-validation technique. A comparative performance analysis of resnet-101 and mobilenet-v3 is performed to validate the efficacy of the deep learning models. The performance of the model with each network is compared. The results demonstrate that the proposed predictive model has accurately classified the foot ulcers in different grades which eventually helps in disease diagnosis and treatment.