TABLE 4.

Mean Accuracy Values and 95% Confidence intervals of the Three Top‐Performing Models Along the 10‐Fold Cross‐Validation for Each Sequence

		Accuracy (mean ± 95% CI) Individual Slice			Accuracy (mean ± 95% CI) Entire Sequence
Sequence	CNN Architecture	Global	Q0 Class	Q1 Class	Global	Q0 Class	Q1 Class
T2WI	VGG11	89.95 ± 0.02	84.16 ± 0.02	96.11 ± 0.04	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
	ResNet101	87.28 ± 0.04	80.37 ± 0.03	94.34 ± 0.05	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
	ResNet50	84.93 ± 0.02	76.42 ± 0.04	93.53 ± 0.01	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
DWI	ResNet152	79.83 ± 0.04	62.13 ± 0.05	97.46 ± 0.03	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
	VGG19‐BN	78.82 ± 0.05	60.08 ± 0.07	97.45 ± 0.03	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
	DenseNet121	75.51 ± 0.05	55.16 ± 0.07	95.78 ± 0.03	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
ADC	DenseNet161	76.64 ± 0.04	64.11 ± 0.06	89.06 ± 0.03	92.31 ± 0.00	83.33 ± 0.00	100.00 ± 0.00
	ResNet50	76.09 ± 0.03	64.11 ± 0.05	87.94 ± 0.02	92.31 ± 0.00	83.33 ± 0.00	100.00 ± 0.00
	VGG13‐BN	73.14 ± 0.05	58.73 ± 0.04	87.51 ± 0.07	92.31 ± 0.00	83.33 ± 0.00	100.00 ± 0.00
DCE	ShuffleNet(v2‐x1‐0)	96.62 ± 0.01	100.00 ± 0.00	93.96 ± 0.01	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
	ShuffleNet(v2‐x0‐5)	95.66 ± 0.04	99.79 ± 0.07	92.15 ± 0.02	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00
	MNASNet1‐0	93.73 ± 0.03	98.90 ± 0.01	93.71 ± 0.05	100.00 ± 0.00	100.00 ± 0.00	100.00 ± 0.00

CNN = convolutional neural networks; Q0 = low‐quality image; Q1 = high‐quality image.