Table 5.

Thyroid classification results with conventional deep learning using pooled external test data.

Deep learning algorithm	Accuracy (%)	Specificity (%)	Sensitivity (%)	PPVa (%)	NPVb (%)	F1 score (%)	AUROC (%)
VGG19	71.0	56.0	86.0	66.2	80.0	74.8	79.3
ResNet50	77.0	72.0	82.0	74.5	80.0	78.1	81.2
ResNext50	80.0	72.0	88.0	75.9	85.7	81.5	89.7
SE-ResNet50	66.0	48.0	84.0	61.8	75.0	71.2	73.4
SE-ResNext50	76.0	58.0	94.0	69.1	90.6	79.7	91.0

^aPPV: positive predictive value.

^bNPV: negative predictive value.