Development of a convolutional neural network (CNN) to detect surgical site infections (SSIs) from wound photographs with high accuracy.
Model training on a large multicenter, multispecialty retrospective wound image dataset (4978 images), with rigorous external validation (407 images from 95 patients) from a large academic center.
Demonstrated model performance with an area under the curve (AUC) of 0.91 internally and 0.82 externally, showing strong discrimination and clinical utility.
Potential for integration into telemedicine workflows to support automated triage and early diagnosis, reducing delays and variability in postoperative SSI detection.
