Abstract
This study aims to develop and internally validate an AI approach based on a deep learning (DL) algorithm to classify wound photographs as either 'healing well' or 'requiring review' after a total knee arthroplasty (TKA).
A prospective cohort study will be conducted at a single, high-volume Elective Orthopaedic Centre. Adult patients who have undergone primary TKAs will be recruited either upon re-attendance at a wound review clinic or if they have a wound concern. Within the first two weeks postoperatively, an orthopaedic research fellow will obtain two standardized photographs of the wound, and the participant will complete a six-item symptom survey. Two blinded consultant orthopaedic surgeons will independently label each case (with the photographic and survey knowledge only) as 'healing well' or 'requires review'. The dataset of wound images will be split into an 80:20 ratio and a pre-trained DL algorithm will be fine-tuned using 80% of the data with five-fold cross-validation being employed. The folds will be generated using stratification not only by outcome labels but also by demographic variables in order to maintain similar distributions across folds. The remaining 20% will be the test set allowing for internal validation and assessment of the efficacy of the developed algorithm. Ethical approval has been granted by the NHS REC/HRA (IRAS 340642).
This study will generate one of the first prospectively validated DL tools for orthopaedic wound triage. Embedding objective imaging and symptom data into a DL algorithm will allow for early detection of complications, timely intervention, streamlined follow-up, and support NHS digital-first pathways. This study's design directly mirrors NHS post-TKA pathways, supporting translatability into the current postoperative workflow for patients. The development of an early-detection system further enables patients to communicate concerns and receive timely assessment and treatment of any postoperative wound issues. The findings of this study will be disseminated through peer-reviewed publications and presentations at national and international conferences.