Table 2. Performance of binary classification DL models.

Datasets	Method	Positives	Negatives	TP	TN	FP	FN	Sensitivity (%)	Specificity (%)	Accuracy (%)	AUC
Internal validation set (n=396)	PM1	238	158	190	134	48	24	88.8	73.6	81.8	0.84
	AM1	229	167	194	147	35	20	90.7	80.8	86.1	0.89
	PM2	225	171	175	132	50	39	81.9	72.5	77.8	0.83
	AM2	213	183	176	145	37	38	82.2	79.7	80.8	0.87
Test set (n=142)	PM1	85	57	59	43	26	14	80.8	62.3	71.2	0.73
	AM1	78	64	65	56	13	8	88.4	80.5	84.5	0.86
	PM2	79	63	57	47	22	16	78.1	68.1	73.2	0.72
	AM2	73	69	58	54	15	15	79.5	78.3	78.9	0.82

DL, deep learning; TP, true positive; TN, true negative; FP, false positive; FN, false negative; AUC, area under the receiver operating characteristic curve; AM1, VGGNet-16-based AT model 1; AM2, ResNet-50-based AT model 2; PM1, pretraining VGGNet-16-based model 1; PM2, pretraining ResNet-50-based model 2; ResNet, residual network; VGGNet, Visual Geometry Group network.