Table 4.

The proposed method has superior performance in detecting pediatric patients with Cerebral Visual Impairment (CVI) from those without CVI using the white matter volume normalized for head size (dataset 4.p) and comparable high performance in detecting adult patients with Alzheimer’s Disease from age-matched controls using the cortical grey matter of the temporal lobe normalized for head size (MIRIAD, dataset 4.a).

	Pediatric dataset 4.p			Adult dataset 4.a
	icobrain-dl	childmetrix	FastSurfer	icobrain-dl	icobrain-nondl	FastSurfer
AUC	0.69	0.48	0.60	0.99	0.98	0.98
Accuracy	0.71	0.57	0.67	0.96	0.93	0.96
Specificity	0.86	0.57	0.86	0.91	0.91	0.96
Sensitivity	0.64	0.57	0.57	1	0.96	0.96

AUC = area under the curve.The AUC obtained by icobrain-dl was significantly higher than the AUC obtained by childmetrix ($p<0.05$) while there was no statistically significant difference between icobrain-dl and FastSurfer, and icobrain-dl and icobrain v5.9 using the DeLong test. The accuracy, specificity, and sensitivity metrics are assessed at the maximum value of the Youden index.