RGB-D Camera-Based Automatic Wound-Measurement System

Chronic wounds can lead to serious complications such as infection and amputation and require effective long-term care and monitoring. However, manual wound measurement is inaccurate and may be painful. The development of a noncontact, low-cost, and accurate automatic wound-measurement system is essential but remains challenging. In this study, we developed an automatic wound-measurement system that can automatically segment wound images and measure the wound area from color (red-green-blue [RGB]) and depth (D) images of the wound. The hardware includes an RGB-D camera, a Linux development board, a touchscreen, and a lithium battery. Based on this hardware, we developed a novel deep learning framework, HarDNet-FSEG, for segmenting wound images, and further proposed edge-based and surface-based methods to measure the area of both flat and curved wounds. Evaluated on two publicly available datasets and a foot ulcer phantom experiment, the average dice score of our wound segmentation method exceeded 0.86, and the accuracy of our wound area measurement method exceeded 95%. The proposed methods outperformed most existing methods for the segmentation and area measurement of wounds. The proposed noncontact, low-cost, and accurate portable wound measurement device will promote the clinical application of automatic wound measurement.