DFUCare: Deep learning platform for diabetic foot ulcer detection, analysis, and monitoring

Abstract

Abstract Diabetic foot ulcers (DFUs) are a severe complication among diabetic patients and often result in amputation and even mortality. Early recognition of infection and ischemia is crucial for improved healing, but current methods are invasive, time-consuming, and expensive. To address this need, we have developed DFUCare, a platform that uses computer vision and deep learning (DL) algorithms to non-invasively localize, classify, and analyze DFUs. The platform uses a combination of CIELAB and YCbCr color space segmentation with a pre-trained YOLOv5s algorithm for wound localization achieving an F1-score of 0.80 and an mAP of 0.861. Using DL algorithms to identify infection and ischemia, we achieved a binary accuracy of 79.76% for infection classification and 94.81% for ischemic classification on a validation set. DFUCare also measures wound size and performs tissue color and textural analysis to allow comparative analysis of macroscopic features of the wound. We tested DFUCare performance in a clinical setting to analyze the DFUs collected using a cell phone camera. DFUCare successfully segmented the skin from the background, localized the wound with less than 10% error, and predicted infection and ischemia with less than 10% error. This innovative approach has the potential to deliver a paradigm shift in diabetic foot care by providing a cost-effective, remote, and convenient healthcare solution.