The system was developed in the Department of Plastic, Hand, and Reconstructive Surgery at the University Hospital Regensburg and in the Clinic for Plastic, Aesthetic, Hand, and Reconstructive Surgery at Caritas Hospital St. Josef. The research team led by Prof. Dr. Dr. Lukas Prantl and Dr. Sally Kempa.

For training the AI, more than 1,000 wound images were used. The system learned to recognize and classify various wound characteristics and to derive diagnostic and therapeutic recommendations. Factors considered include wound depth and cause, degree of inflammation, tissue quality, necrotic cells, and wound extent.

Explainable AI instead of Black Box

A focus of the development is on the traceability of AI results. For this, the method Grad-CAM is used. It highlights image areas in color that were particularly relevant for the decision of the AI system.

Doctors can thereby trace on what basis a diagnosis or therapy recommendation was made. Thus, the AI is designed as an assistance system to support medical decision-making.

AI also monitors healing progress

The system is intended not only to help with the initial assessment of a wound. With the help of follow-up images, it can also document the wound and healing progress , recognize changes and thus support the adaptation of the therapy. The scientific results of the Regensburg working group have already been published in the journal Diagnostics.

In perspective, the approach could also be used for burns and other complex soft tissue injuries. The researchers aim to improve wound healing, detect complications earlier, and ideally avoid amputations.

Perspective for telemedicine care

In the long term, the researchers also see potential for telemedicine applications. Especially in regions with limited access to specialized wound centers, AI could support medical staff in assessing wounds.

The development from Regensburg thus shows how explainable Artificial Intelligence can be used in medicine: not as a replacement for medical decisions, but as a comprehensible tool to support diagnostics and therapy.