Table 3.

Comparison of the test performance of imputation methods with respect to volumetric features extracted from whole-brain imputed images. Results were averaged across the 10 test sets and averaged across all cortical regions defined by the Desikan-Killiany-Tourville atla. Lower error rate indicates better performance. The values in () are differences between the observed progression rate (using observed images at both time points) and the estimated progression rate using the imputed image at the latter time point. Lower difference means better performance. P indicates past image. PF indicates past and following images. The best test results across all methods and models are bolded.

Models	Error Rate (ANTs)	Error Rate (FreeSurfer)	Progression Rate (ANTs)	Progression Rate (FreeSurfer)
Naive-P	0.0310 ± 0.0028	0.1877 ± 0.0033	0	0
			(–)	(–)
AE-P	0.0284 ± 0.0002	0.1078 ± 0.0004	0.0422 ± 0.0038	0.1358 ± 0.0030
			(−0.0121)	(−0.1328)
ReMiND-P	0.0228 ± 0.0003	0.0892 ± 0.0029	0.0361 ± 0.0033	0.1024 ± 0.0004
			(−0.006)	(−0.051)
Naïve-PF	0.0218 ± 0.0009	0.1393 ± 0.0032	0.0365 ± 0.0012	0.1842 ± 0.0031
			(−0.0157)	(−0.1332)
AE-PF	0.0192 ± 0.0008	0.0937 ± 0.0019	0.0231 ± 0.0009	0.0949 ± 0.0013
			(−0.0023)	(−0.0439)
ReMiND-PF	0.0178 ± 0.0002	0.0509 ± 0.0003	0.0226 ± 0.0006	0.0872 ± 0.0018
			(−0.0017)	(−0.0362)