Table 3.

Measures of accuracy in the brain age prediction using the selected retrained CNN (VGG16-based DeepBrainNet) with the linear correction layer for the MRIs for each modality in the testing held-out sample.

Modality	Testing Sample MAE		Correlation		${\mathrm{R}}^2$		Within-subject Difference with original MPRAGE		$\widehat{\text{MAE}}$
	Mean	95% CI	Mean	95% CI	Mean	95% CI	Mean	95% CI	Mean	95% CI
MPRAGE	5.33	[4.70, 6.03]	0.96	[0.93, 0.97]	0.90	[0.87, 0.93]			5.28	[4.42, 6.14]
Synthetic MRIs
MPRAGE	4.67	[4.10, 5.41]	0.95	[0.92, 0.96]	0.89	[0.84, 0.92]	−0.57	[−1.5, 0.36]	4.71	[3.87, 5.56]
T1w	5.74	[4.50, 6.03]	0.93	[0.91, 0.95]	0.83	[0.75, 0.88]	0.57	[−0.41, 1.56]	5.86	[5.04, 6.68]
T2w	4.99	[4.48, 5.54]	0.94	[0.92, 0.96]	0.89	[0.85, 0.91]	−0.33	[−1.14, 0.48]	4.95	[4.30, 5.60]
T1wFLAIR	4.81	[3.95, 5.69]	0.96	[0.94, 0.97]	0.89	[0.85, 0.93]	0.14	[−0.98, 1.26]	5.42	[4.40, 6.44]
T2wFLAIR	5.05	[4.59, 5.67]	0.95	[0.94, 0.96]	0.90	[0.87, 0.92]	−0.17	[−1.05, 0.71]	5.12	[4.38, 5.85]
IR	6.47	[5.61, 7.75]	0.91	[0.83, 0.94]	0.81	[0.72, 0.88]	0.91	[−0.19, 2.00]	6.19	[5.16, 7.22]
T2wGRE	18.42	[17.18, 19.66]	0.93	[0.90, 0.94]	−0.99	[−1.51, −0.50]	13.31	[12.35, 14.27]	18.59	[17.77, 19.41]
All synthetic MRIs except T2wGRE	5.21	[4.86, 5.94]	0.94	[0.92, 0.95]	0.88	[0.84, 0.91]	1.98	[1.25, 2.71]	5.37	[4.80, 5.95]

Note. The first, second and third columns are the bootstrapped MAE, Correlation and coefficient of determination of the $correctedbrainage=chronologicalage$ linear model, respectively. The fourth column is the estimated within-subject difference in the corrected brain-PAD between each modality and the original MPRAGE. The fifth column is the “population” estimated MAE.