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

Table 3 Performance comparison between Resnet18 and radiologists without model assistance in the internal-test

	AUC (95% CI)	Accuracy	Sensitivity	Specificity	PPV	NPV	F1	Kappa
Senior
Radiologist C	0.776(0.717,0.835)	0.797	0.727	0.825	0.625	0.883	0.672	0.526
Radiologist D	0.772(0.713,0.832)	0.760	0.800	0.745	0.557	0.903	0.657	0.482
Intermediate
Radiologist E	0.734(0.672,0.797)	0.745	0.709	0.759	0.542	0.867	0.614	0.429
Radiologist F	0.745(0.683,0.807)	0.760	0.709	0.781	0.565	0.870	0.629	0.455
Junior
Radiologist G	0.591(0.521,0.660)	0.734	0.255	0.927	0.583	0.756	0.354	0.218
Radiologist H	0.616(0.547,0.685)	0.693	0.436	0.796	0.462	0.779	0.449	0.236
Resnet 18	0.947(0.915,0.979)	0.885	0.782	0.927	0.811	0.914	0.796	0.717

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

ISSN: 1471-2407