Standardising Wound Image Acquisition Through Edge AI

The high prevalence of chronic wounds, along with their impact on the patient’s life quality and healthcare systems, make them a relevant public health issue and have motivated the rise of digital health solutions to support wound monitoring. This paper presents a new framework that leverages deep learning models to automate wound image acquisition in real-time while guaranteeing focus and inclusion of an adequate periwound area. Considering an adhesive marker as a metric reference, a RetinaNet detection model is responsible for locating the wound and marker regions, further analysed by a post-processing module that validates if both structures are present and verifies if a periwound radius between 4 to 8 centimetres is included. The initial validation of this pipeline demonstrated that the developed algorithms exhibit a robust detection performance for varying acquisition conditions (translated into mAP@.75IOU values of 0.39 and 0.95 for wound and marker detection) and the deployability of the framework in an easily usable mobile application, without causing any performance hindrances. The proposed solution was then tested in a real environment by integrating the whole framework into a mobile application available in Android and iOS. During a two-month pilot study, healthcare professionals tested this application in their clinical practice. According to their feedback, the usage of the mobile application improved image quality and standardisation as the main advantages of the application, confirming the potential of the presented framework to streamline the image acquisition flow and make wound monitoring more reproducible.