Accurate assessment of burn extent and depth is crucial but challenging. This study developed two deep-learning algorithms: one to identify burns and another to classify the need for surgery, evaluating performance across different Fitzpatrick skin types. Using 1105 burn images and 536 background images, the wound identifier algorithm achieved 87.2% accuracy in identifying wound areas, while the wound classifier algorithm showed an AUC of 0.885. The wound identifier performed better on darker skin types, and the wound classifier on lighter skin types. The study concludes that image-based algorithms can support burn assessment, but larger and more diverse datasets are needed.

Constance Boissin, Lucie Laflamme, Jian Fransén, Mikael Lundin, Fredrik Huss, Lee Wallis, Nikki Allorto, Johan Lundin