Application of deep learning in wound size measurement using fingernail as the reference

Table 2 PCCs and Mean euclidean distances obtained by Mask R-CNN with different confidence thresholds

Confidence threshold	mAP@0.5⁽¹⁾	PCC⁽²⁾	Mean Euclidean distance (in pixels)
Confidence threshold	mAP@0.5⁽¹⁾	PCC⁽²⁾	Left keypoints	Right keypoints
0.50	0.977	0.939	528.339	526.048
0.55	0.977	0.939	526.390	524.095
0.60	0.977	0.939	525.925	523.778
0.65	0.977	0.939	526.357	524.114
0.70	0.977	0.939	527.932	525.830
0.75	0.977	0.939	522.985	521.374
0.80	0.977	0.939	518.635	516.536
0.85	0.968	0.938	515.860	514.403
0.90	0.968	0.938	517.506	515.617
0.95	0.968	0.938	494.951	492.504

⁽¹⁾ mAP@0.5: mean Average Precision at an IoU threshold of 0.5. ⁽²⁾ PCC: Pearson correlation coefficient
The best Confidence threshold to obtain the best mAP@0.5, PCC, Mean Euclidean distance is bold

ISSN: 1472-6947