Three healthcare institutions in Western Switzerland provided high-resolution images, short videos, and 3D scans, along with structured clinical metadata, to train an AI model on roughly 4,000 wounds.
© Stefanelli et al.
The AI-based wound segmentation model, developed using the Deeplabv3+ architecture with a ResNet50 backbone, achieved a DICE score of 92% and an Intersection-over-Union (IoU) score of 85%. These scores indicate how accurately the AI model can automatically detect the precise wound area, with an average processing time of 0.3 seconds.
