Table 2.

Performance for binary classification of cannulation difficulty.

Model	VGG		ResNet		DenseNet
	Easy	Difficult	Easy	Difficult	Easy	Difficult
Precision	0.787 ± 0.081	0.434 ± 0.083	0.802 ± 0.047	0.507 ± 0.038	0.772 ± 0.076	0.436 ± 0.047
Recall	0.607 ± 0.048	0.656 ± 0.102	0.719 ± 0.081	0.611 ± 0.098	0.631 ± 0.057	0.615 ± 0.045
F1 score	0.680 ± 0.016	0.511 ± 0.048	0.757 ± 0.062	0.553 ± 0.062	0.694 ± 0.061	0.507 ± 0.026
ACC	0.614 ± 0.024		0.695 ± 0.042		0.627 ± 0.037
AUC	0.657 ± 0.061		0.680 ± 0.021		0.626 ± 0.034

Bold fonts represent the best performance among the methods.

AUC area under the receiver operating characteristic curve, ACC accuracy.