Identification of Alterations in Surgical Wounds Through the Application of Artificial Intelligence in Digital Images

Globally, surgical site infections are common complications that are both serious and costly. While telemedicine has enhanced the remote assessment of surgical wounds, it still faces limitations. This study introduces a convolutional neural network (CNN) model designed to automatically classify digital images of surgical wounds as either altered or unaltered. The study utilized a dataset of 4,262 segmented and expert-labeled images. The CNN model achieved an accuracy of 83.46%, a sensitivity of 81.54%, and an AUROC of 92.22%. Although the MobileNet model demonstrated acceptable performance, it was less effective in comparison. The findings s uggest t hat C NNs a re e ffective for classifying images of surgical wounds, with potential for further improvement using advanced techniques and a multidisciplinary expert panel.