Researchers at the Mayo Clinic have developed an artificial intelligence (AI) system that analyzes patient-submitted photos of postoperative wounds to accurately detect surgical site infections (SSIs).
Study: Imaging Based Surgical Site Infection Detection Using Artificial Intelligence. Image Credit: Alexander Supertramp/Shutterstock.com
The two-stage model identifies surgical incisions with 94 % accuracy and assesses infection risk with an 81 % area under the curve (AUC), offering a scalable option for remote monitoring. This technology could help streamline postoperative care, reduce diagnostic delays, and improve outcomes, especially for patients in outpatient or rural settings.
Background
SSIs are one of the most common postoperative complications, affecting up to 5 % of surgical patients and significantly increasing healthcare costs. Traditionally, detection relies on in-person follow-ups, which can delay diagnosis, particularly for patients living far from care centers. As telehealth becomes more common and outpatient procedures rise, there’s a clear need for reliable remote tools to evaluate wound healing.
To meet this need, Mayo Clinic researchers created an AI system trained on more than 20,000 images from over 6000 patients across nine hospitals. Published in the Annals of Surgery, the model uses a vision transformer architecture to first locate surgical incisions in images and then evaluate them for signs of infection. Unlike manual photo reviews, the AI offers fast and consistent assessments, allowing for earlier interventions and easing clinicians’ workloads. This system may be especially beneficial in underserved areas with limited access to surgical specialists.
AI Model Development and Performance
The AI tool works through a two-stage deep learning process. Patients upload photos of their surgical wounds via an online portal. The first stage identifies whether the image includes a surgical incision, achieving 94 % accuracy. The second stage assesses the incision for infection, with an 81 % AUC—an indicator of strong diagnostic performance.
The model was built on a diverse image dataset to ensure it performs reliably across different skin tones, lighting conditions, and wound types. This approach helps reduce the risk of algorithmic bias, a key concern in healthcare AI. Researchers chose a vision transformer architecture due to its ability to capture complex visual patterns, outperforming traditional convolutional neural networks in early testing.
Dr. Cornelius Thiels, co-senior author of the study, noted that the tool is designed to support—not replace—clinicians. “This tool doesn’t replace doctors but helps prioritize cases needing urgent attention,” he said. Looking ahead, the team hopes future versions may detect infection signs even before they become visible to the human eye.
Clinical Implications and Next Steps
This technology has the potential to reshape postoperative care by enabling real-time, remote wound monitoring. For patients, that means faster reassurance—or quicker escalation if problems arise—ultimately reducing anxiety and preventing complications. For providers, it offers a way to focus attention on high-risk cases, an especially valuable feature in settings with limited resources.
Dr. Hala Muaddi, the study’s first author, emphasized the tool’s role in expanding access to care: “As outpatient surgeries increase, this technology ensures patients aren’t left without follow-up.” The research team is now conducting prospective studies to explore how the system can be integrated into routine clinical workflows.
The project is supported by Dalio Philanthropies and the Simons Foundation, with ongoing work focused on enhancements like multi-language support and a dedicated smartphone app. If successfully validated, this tool could help reduce readmissions and lower healthcare costs, setting a new benchmark for remote postoperative care.
Conclusion
This AI-based system marks a promising step forward in surgical site infection detection, offering timely, accurate, and consistent remote monitoring. By automating the assessment of wound images, it addresses key challenges in telehealth, particularly for patients in rural or underserved communities. While additional validation is underway, the early results suggest it could become a valuable addition to postoperative protocols, improving outcomes while easing clinician burden.
Journal Reference
Hala Muaddi, Choudhary, A., Lee, F., Anderson, S., Habermann, E., Etzioni, D., McLaughlin, S., Kendrick, M., Hojjat Salehinejad, & Thiels, C. (2025). Imaging Based Surgical Site Infection Detection Using Artificial Intelligence. Annals of Surgery. DOI: 10.1097/sla.0000000000006826. https://journals.lww.com/annalsofsurgery/abstract/9900/imaging_based_surgical_site_infection_detection.1357.aspx
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.