Oral epithelial dysplasia detection and grading in oral leukoplakia using deep learning

BMC Oral Health

Table 3 The performances of convolutional neural networks

	Accuracy (%)	95% CI (%)	AUC	95% CI
ShuffleNet-V2
224 px & 10×	91.1	(90.5, 91.8)	0.957	(0.952, 0.962)
224 px & 20×	89.0	(88.7, 89.4)	0.957	(0.955, 0.959)
512 px & 10×	72.6	(70.3, 75.0)	0.729	(0.692, 0.763)
512 px & 20×	80.8	(79.8, 81.8)	0.809	(0.794, 0.823)
ResNet-50
224 px & 10×	90.6	(89.9, 91.3)	0.954	(0.949, 0.960)
224 px & 20×	88.0	(87.6, 88.4)	0.930	(0.926, 0.933)
512 px & 10×	72.6	(70.3, 75.0)	0.633	(0.598, 0.666)
512 px & 20×	85.1	(84.2, 86.0)	0.894	(0.884, 0.903)
Inception-V4
224 px & 10×	93.8	(93.3, 94.4)	0.979	(0.976, 0.983)
224 px & 20×	94.9	(94.6, 95.1)	0.987	(0.986, 0.988)
512 px & 10×	75.2	(73.0, 77.5)	0.603	(0.565, 0.640)
512 px & 20×	82.5	(81.5, 83.5)	0.772	(0.755, 0.788)
EfficientNet-B0
224 px & 10×	92.8	(92.3, 93.4)	0.966	(0.961, 0.971)
224 px & 20×	97.5	(97.3, 97.6)	0.993	(0.992, 0.994)
512 px & 10×	76.6	(74.4, 78.9)	0.741	(0.709, 0.774)
512 px & 20×	89.6	(88.8, 90.4)	0.928	(0.919, 0.937)

CI: confidence interval; AUC: area under the receiver operating characteristic curve; px: pixel

ISSN: 1472-6831