Table 5.

Summary of papers on chronic wound detection.

Reference	Subject and Classes (Each Bullet Represents a Different Model)	Methodology	Original Images (Training Samples)	Results
Goyal et al. (2018) [65]	Diabetic foot ulcer Wound/Non-Wound	Cropping and resizing Augmentation was applied Faster R-CNN based on InceptionV2 was used for object detection	Normal: 2028 (28,392)Abnormal: 2080 (29,120)	mAP: 0.918
Amin et al. (2020) [42]	Diabetic foot ulcer Ischemia/Non-Ischemia Infection/Non-Infection	Objects were detected with the YOLOv2 algorithm. Detected objects were classified with a 16-layer CNN.	Ischemia: 9870 (9870) Infection: 5892 (5892)	Ischemia: Accuracy: 0.97 mAP: 0.95 Infection: Accuracy: 0.99 mAP: 0.9
Han et al. (2020) [87]	Diabetic foot ulcer Wagner Grades 0–5	K-means++ was used to select anchor box sizes based on the training dataset Fast R-CNN was used for detection and classification, and selected anchor boxes were used for region proposals	2688 (2668)	mAP: 0.9136
Anisuzzaman et al. (2022) [25]	Wound Wound/Non-Wound	Augmentation was applied YOLOv3 was used for wound detection	1800 (9580)	F1: 0.949 mAP: 0.973
Huang et al. (2022) [88]	Wound Blocked Blood Vessel/Suture/Ulceration	Augmentation was applied Fast R-CNN using the ResNet101 backbone was used for wound localization and classification The GrabCut and SURF algorithms were applied for wound boundary refinement	727 (3600)	Accuracy: 0.88 mAP: 0.87