Back to 2026 Abstracts
Comparison Of A Novel Deep Learning Model With Healthcare Professionals In Chronic Wound Infection Detection
Bilal M. Chaudhry, MD1, Palawat Busaranuvong, PhD
2, Francisco Batiz-Fabella, MD
3, Eddy C. Rios, BS
3, Emma N. Mastro, BS
3, Diane Strong, PhD
2, Bengisu Tulu, PhD
2, Emmanuel Agu, PhD
2, Raymond M. Dunn, MD
4, Giorgio Giatsidis, M.D Ph.D
3.
1Rutgers University/ RWJ Barnabas Health, Toms River, NJ, USA,
2Worcester Polytechnic Institute, Worcester, MA, USA,
3University of Massachusetts Chan Medical School, Worcester, MA, USA,
4University of Massachusetts Chan Medical Center, Worcester, MA, USA.
PURPOSE:Chronic wounds affect 10.5 million Medicare beneficiaries. With an aging population demand for wound expertise is increasing. SCARWID (Synthetic Caption Augmented Retrieval for Wound Infection Detection), a Deep Learning (DL) model, was created to detect infections in areas without wound expertise. We evaluated SCARWID performance and compared it to healthcare professionals (HCP).
METHODS:1,208 Diabetic Foot Ulcer (DFU) images with expert assigned labels (infected vs non-infected) were used. SCARWID was trained and evaluated with 5-fold cross-validation. The same images were divided among HCPs—medical students (n=4), plastic surgery residents (n=5), emergency physicians (n=3), and plastic surgeons (n=4). Accuracy, sensitivity (SN), specificity (SP), and positive predictive value (PPV) were calculated. Paired accuracy differences between each HCP group and SCARWID were estimated.
RESULTS:SCARWID achieved 81% accuracy, 85% SN, 78% SP, and 79% PPV. HCP overall reached 64% accuracy, 52% SN, 74% SP, and 72% PPV. Students averaged 68% accuracy (SN 68%, SP 70%, PPV 70%); residents 63% accuracy (SN 35%, SP 89%, PPV 85%); emergency physicians 57% accuracy (SN 64%, SP 50%, PPV 54%); and plastic surgeons 65% accuracy (SN 48%, SP 80%, PPV 72%). Paired accuracy differences (group − SCARWID) were −0.134 for students (−0.284 to +0.016; p=0.06), −0.188 for residents (−0.276 to −0.101; p=0.003), −0.245 for emergency physicians (−0.412 to −0.078; p=0.02), and −0.167 for plastic surgeons (−0.290 to −0.045; p=0.02).
CONCLUSION:Overall, SCARWID outperformed every HCP group in accuracy of infection detection. These results support the potential of DL to augment infection triage in a remote setting.
Back to 2026 Abstracts
