Table 2.

The classification results of ResNet18, VGG16 and ResNet50 at slice level

	Delong test	AUC	Accuracy	Sensitivity	Specificity
ResNet18	Reference	0.86	80.28% (3073/3828)	73.28% (1445/1972)	87.72% (1628/1856)
VGG16	< 0.0001	0.67	62.77% (2403/3828)	77.43% (1527/1972)	47.20% (876/1856)
ResNet50	< 0.0001	0.74	67.40% (2580/3828)	52.23% (1030/1972)	83.51% (1550/1856)

For the slice level, the sensitivity and specificity refer to the ratio of successfully identifying the image slices with GABAB receptor antibody encephalitis and that successfully identifying the image slices with LGI1 antibody encephalitis, respectively