Imaging-based Surgical Site Infection Detection Using Artificial Intelligence

Area of Science:

• Medical Artificial Intelligence
• Digital Health
• Surgical Site Infection Detection

Background:

• Increasing outpatient surgeries and remote monitoring generate significant administrative tasks for clinicians.
• Early detection of surgical site infections (SSIs) is critical for minimizing postoperative complications.
• Patient-submitted wound images via online portals are becoming more common.

Purpose of the Study:

• To create an artificial intelligence (AI)-based system for evaluating and prioritizing patient-submitted postoperative wound images.
• To automate the assessment of wound images and detect potential surgical site infections (SSIs).

Main Methods:

• Developed a two-stage AI model for incision detection and SSI detection in postoperative wound images.
• Utilized a dataset of 6060 patient images from Mayo Clinic hospitals (2019-2022), including SSI outcomes from the National Surgical Quality Improvement Program (NSQIP).
• Evaluated four pretrained architectures using cross-validation, data augmentation, and image quality assessment, including sensitivity analysis across racial groups.

Main Results:

• The Vision Transformer model achieved high accuracy in incision detection (0.94) and SSI detection (0.73).
• The AI pipeline demonstrated robust performance in image quality assessment.
• Performance was consistent across different racial subgroups, indicating equitable effectiveness.

Conclusions:

• The developed AI pipeline shows significant potential for automating the assessment of postoperative wound images.
• This technology can effectively aid in the early detection of surgical site infections (SSIs).
• The AI system promises to reduce clinician workload and enhance the quality of postoperative patient care.