Table 2.

Summary of deep learning in object detection for images of DFU.

References	Purpose	Network structure	Contributions	Limitations	Results
Da Costa et al., 2021 (41)	DFU detection	•Adaptive faster R-CNN	•Better performance •Improving the accuracy of detecting small lesions	•Slower speed	•Precision: 91.4% •F1-score: 94.8%
Goyal et al., 2019 (8)	Detection and localization of DFU on mobile devices	•Faster R-CNN with InceptionV2 •Two-tier transfer learning	•Better performance •More accurate •Lightweight •Reducing computation •Decreasing internal covariate shift •Improving convergence	•Worse than R-FCNResnet101	•Precision: 91.8% •48 ms per image
Han et al., 2020 (44)	Real-time detection and location for the Wagner grades of DFUs	•Refined YOLO v3 •On smartphones	•Single-stage •Better acquisition of object features •Improving accuracy	•Inter-class similarity	•Accuracy:91.95% •Outperformed mAP •Good trade-off
Goyal et al., 2020 (45)	DFU detection	•Refined EfficientDet with distinct bounding boxes	•A weighted bi-directional feature pyramid network •Uniform scale •Minimizing false positives and false negatives	•No own data	•Without a report
Yap et al., 2020 ( 13)	DFU detection	•An ensemble model	•A comprehensive evaluation •A variant of faster R-CNN with the best performance	•High false positives rate •Difficult to discriminate from other skin	•mAP: 0.6940 •F1-Score: 0.7434