Non-contact chronic wound analysis using deep learning

Abstract

The management of chronic wounds requires an analysis of the causative factors, an initial assessment of the wound, and the continuous monitoring of the wound. Wound specialists rely on images for diagnosis and treatment. Objective wound measurements are essential for the effective management of the wound. In this study, we applied deep learning algorithms to segment wound area from 2D images. We employed the public WoundsDB dataset, which contains wound images of 47 patients. Using the U-Net architecture with the EfficientNet-B2 encoder, we achieved an average IOU of 0.8674. Our study provides a step towards the automated analysis of wound that could help physicians to measure and track the progression of wound treatment in the clinic.