1891-LB: DFUCare—SAM powered Segmentation for Enhanced Diabetic Foot Ulcer Analysis Platform

Introduction and Objective: Diabetic foot ulcers (DFUs) require precise assessment for effective treatment. We previously introduced a deep learning-based platform that provides DFU detection, analysis, and monitoring. We enhanced it by integrating the Segment Anything Model (SAM) for wound segmentation, aiming to improve wound feature analysis.

Methods: DFU images from DFUC2020 (n=2000) and real-world clinical settings (PGIMER, India, n=643) were used. The ground truths are generated by healthcare professionals and trained researchers. YOLO was used to detect wound regions, and SAM was employed to segment them. The DFUCare platform extracted wound size, color, and roughness information, while classifying infection and ischemia.

Results: In a 10-fold cross-validation, YOLO yielded a mean average precision of 0.8375, and SAM achieved a median IoU of 0.8096, indicating robust and consistent wound delineation. DFUCare accurately extracted wound characteristics and classified infection (81.2%) and ischemia (95.4%). Preliminary tests on the PGIMER data confirmed the robustness of the detection and segmentation steps in real-world clinical scenarios.

Conclusion: By incorporating a segmentation stage, DFUCare now enables more precise wound size estimation, roughness quantification, and color assessment compared to the localization-only approach. Its improved accuracy and real-world usability offer a cost-effective, remote-friendly solution for DFU care.

D. Choi: None. C. Kharbanda: None. K. Saju: None. S.K. Bhadada: None. M.K. Bhasin: None.