Chronic Single and Multiple Diabetic, Pressure and Venous Ulcers Detection Using YOLO Networks

Chronic wounds are one of the most emerging health problems in the world. Globally, around 2 out of 1000 people are estimated to suffer from chronic wounds. Proper assessment is a key factor in wound management because every wound requires a different treatment plan. Therefore, the need for improved assessment technology for these types of wounds has been constantly raised. For the past few years, the usage of artificial intelligence (AI) technology has tremendously increased, especially in the field of medical diagnosis. This study aims to detect and classify chronic wounds using YOLOv7 and YOLOv8 and compare the results. For the dataset, the AZH dataset, which is collected by the AZH (advancing the zenith of healthcare) wound and vascular center in Milwaukee, Wisconsin, was used. Images of various chronic wounds, like venous and diabetic ulcers, are included in this dataset. With a ratio of 70:15:15, 987 photos were split into training, validation, and testing categories. Data augmentation was also performed, producing 2,073 images as a training dataset. YOLOv7 and YOLOv8 models were developed, and YOLOv8 outperformed YOLOv7 in every aspect, including mAP and prediction accuracy. Hence, the output of this study suggests the potential application of YOLOv8 in developing devices that detect wounds based on computer vision.