Establishing an AI Pressure Injury Database to Improve Clinical Nursing Efficiency.

Treatment varies depending on the condition and stage of the pressure injury. Therefore, there is a need for a friendly non-invasive assessment method for pressure injuries, as well as regular assessment and monitoring of the extent of the injury. This study used over 20,000 pressure injury images collected in the the Comprehensive Medical Data Center. Staging assessment and marking were performed according to the definitions of the NPIAP staging system by senior nurses. We randomly divided the photos into a training set and a validation set and developed the model through deep learning methods. Using Swin transformer model in the identification of S1-S4, the F1 score of the identification accuracy is about 0.6-0.8, but for X and DTI, the accuracy is still low.