Research question Can we successfully address patient and clinician feedback from the WISDOM feasibility study by improving the sensitivity of the AI model to identify images of surgical wounds with infection to at least 90% and wound healing concerns to at least 95%, and can we improve the sensitivity of the AI model to identify infection indicators in dark skin tones (compared against the WISDOM AI model)? Background Around 5% of patients develop an infection in their wound after surgery. These can result in re-admissions, further surgery and cost the NHS an estimated £1billion to treat. Detecting infections early can reduce their severity making them easier and cheaper to treat, however wound monitoring is not widely carried out. Digital surgical wound monitoring using photos taken by patients can be implemented easily but it creates an additional workload for clinicians. Using AI to identify images with infections can help clinicians manage this additional workload. The WISDOM (NIHR204508) feasibility study developed and tested an AI surgical wound monitoring platform that was liked by clinicians and patients but a higher sensitivity for identifying infections was preferred. The AI model also showed lower sensitivity for identifying discolouration (a sign of infection) in darker skin tones. The James Lind Alliance reports identifying infections early as a top priority. Aims To further train the AI model to identify images of surgical wounds with signs of infection to at least 90%, and wound complications to 95%. To further train the AI model to increase the sensitivity to identify infection indicators in dark skin tones when compared against the original WISDOM AI model. Objectives are to deliver a validated improved AI model, an updated health economic (HE) assessment and an updated plan for adoption into the NHS, ready to be used within a subsequent trial. Methods We will add 42,000 new images, of which 11,000 are annotated with patient skin tone data, to our existing dataset of 28,000 surgical wound images. Using this dataset we will 1) retrain existing models using additional data, 2) add pre-screening object detection and allow models to run at a more aggressive operating point and 3) leverage transfer learning to apply existing methods to newer base models. This will deliver an improved AI model ready for validation. The model will be assessed, using a separate testing dataset, for sensitivity to identify images of wounds with an infection or wounds with a healing concern. Performance across light and dark skin tones will also be assessed. Timelines Improved AI model development completed by month 4 and fully validated by month 7. Revised implementation and adoption plan, including updated HE assessment will be completed by month 7. Impact We will deliver; • An improved AI model that identifies images of surgical wounds that are infected or have a healing complication, across all skin tones • An updated HE assessment of the costs and benefit of the model to the NHS • An updated adoption strategy into the NHS Dissemination activities include two workshops, website, social media and publications. Increased acceptability of AI in digital surgical wound monitoring, plus improved identification in patients with darker skin tones, will result in improved clinical outcomes through early infection detection and cost savings for the NHS. Inclusion EDI input to study aims, data analysis, design and dissemination.