Deep learning-assisted diagnosis of benign and malignant parotid tumors based on ultrasound: a retrospective study

Table 2 The performance comparison of different models in internal- and external- test set

Model		AUC (95% CI)	Accuracy	Sensitivity	Specificity	PPV	NPV	F1	Kappa
Resnet 18	Internal-test	0.947(0.915,0.979)	0.885(0.840, 0.930)	0.782(0.673, 0.891)	0.927(0.883, 0.971)	0.811(0.706, 0.917)	0.914(0.867, 0.960)	0.796	0.717
	External-test	0.925(0.857, 0.992)	0.898(0.821, 0.975)	0.833(0.535, 1.132)	0.906(0.827, 0.984)	0.500(0.190, 0.810)	0.980(0.940, 1.019)	0.625	0.570
Resnet 50	Internal-test	0.908(0.867,0.979)	0.797(0.740, 0.854)	0.491(0.359, 0.623)	0.920(0.874, 0.965)	0.711(0.566, 0.855)	0.818(0.757, 0.879)	0.581	0.452
	External-test	0.896(0.818, 0.974)	0.864(0.777, 0.952)	0.833(0.535, 1.132)	0.868(0.777, 0.959)	0.417(0.138, 0.696)	0.979(0.937, 1.020)	0.556	0.486
Vgg 11	Internal-test	0.902(0.860,0.944)	0.839(0.786, 0.891)	0.691(0.569, 0.813)	0.898(0.847, 0.949)	0.731(0.610, 0.851)	0.879(0.824, 0.933)	0.710	0.598
	External-test	0.887(0.806, 0.968)	0.847(0.756, 0.939)	0.833(0.535, 1.132)	0.849(0.753, 0.945)	0.385(0.120, 0.649)	0.978(0.936, 1.020)	0.526	0.450
Vgg 16	Internal-test	0.907(0.866,0.948)	0.839(0.786, 0.891)	0.764(0.651, 0.876)	0.869(0.812, 0.925)	0.700(0.584, 0.816)	0.902(0.851, 0.952)	0.730	0.616
	External-test	0.887(0.806, 0.968)	0.814(0.714, 0.913)	0.833(0.535, 1.132)	0.811(0.706, 0.917)	0.333(0.095, 0.572)	0.977(0.933, 1.021)	0.476	0.387
Mobilenetv2	Internal-test	0.878(0.832,0.925)	0.839(0.786, 0.891)	0.709(0.589, 0.829)	0.891(0.838, 0.943)	0.722(0.603, 0.842)	0.884(0.831, 0.937)	0.716	0.603
	External-test	0.858(0.770, 0.947)	0.797(0.694, 0.899)	0.667(0.289, 1.044)	0.811(0.706, 0.917)	0.286(0.049, 0.522)	0.956(0.895, 1.016)	0.400	0.300

AUC area under the curve, PPV positive prediction value, NPV negative prediction value

ISSN: 1471-2407