Table 4.

Ablation study on the 3D classification network.

Task and model			Size (MB)		Accuracy, mean (SD)		Recall, mean (SD)		Specificity, mean (SD)		Precision, mean (SD)		F1-score, mean (SD)		AUROCa, mean (SD)		P value
1
	Proposed	14.2		0.878 (0.017)		0.853 (0.032)		0.913 (0.067)		0.933 (0.045)		0.89 (0.012)		0.00959 (0.00011)		N/Ab
	Model 1	4.0		0.858 (0.03)		0.827 (0.046)		0.901 (0.058)		0.921 (0.043)		0.87 (0.028)		0.00947 (0.00019)		<.001
	Model 2	34.5		0.884 (0.014)		0.862 (0.021)		0.914 (0.041)		0.933 (0.03)		0.895 (0.012)		0.00961 (0.00009)		.26
	Model 3	64.8		0.864 (0.022)		0.851 (0.062)		0.887 (0.074)		0.914 (0.053)		0.878 (0.019)		0.0095 (0.00019)		.003
2
	Proposed	14.2		0.783 (4.0)		0.808 (0.025)		0.77 (0.054)		0.652 (0.06)		0.721 (0.042)		0.00853 (0.0003)		N/A
	Model 1	4.0		0.758 (0.048)		0.712 (0.118)		0.783 (0.065)		0.636 (0.064)		0.668 (0.075)		0.00817 (0.0006)		.006
	Model 2	34.5		0.782 (0.036)		0.795 (0.071)		0.775 (0.074)		0.659 (0.071)		0.716 (0.032)		0.00862 (0.00031)		.91
	Model 3	64.8		0.756 (0.056)		0.76 (0.059)		0.754 (0.109)		0.634 (0.088)		0.685 (0.037)		0.00826 (0.000047)		.003

^aAUROC: area under the receiver operating characteristic curve.

^bN/A: not applicable.