Insights into Imaging

Table 2 Voxel-wise z-score root-mean-square error (RMSE) values after veganbagel processing of Alzheimer’s disease patients and comparing the results after defacing with the respective z-scores derived from the respective full face 3D T1 imaging

From: Impact of defacing on automated brain atrophy estimation

	Unaccelerated imaging					Accelerated imaging
	Failed processing^a	Mean RMSE ± SD	Range (IQR)	Outliers		Failed processing^a	Mean RMSE ± SD	Range (IQR)	Outliers
	Failed processing^a	Mean RMSE ± SD	Range (IQR)	Grubbs’s test	Benchmark^b	Failed processing^a	Mean RMSE ± SD	Range (IQR)	Grubbs’s test	Benchmark^b
afni_refacer	2/154	0.45 ± 0.97	0.11–5.84 (0.24–2.16)	16 (10.5%)	18 (11.8%)	0/114	0.28 ± 0.26	0.12–2.07 (0.26–0.65)	13 (11.4%)	16 (14%)
fsl_deface	0/154	0.09 ± 0.08	0.02–0.50 (0.10–0.24)	10 (6.5%)	4 (2.6%)	0/114	0.23 ± 0.47	0.03–2.49 (0.16–1.06)	21 (18.4%)	18 (15.8%)
mri_deface	32/154	0.20 ± 0.35	0.02–1.92 (0.11–1.15)	25 (20.5%)	15 (12.3%)	30/114	0.30 ± 0.57	0.03–2.44 (0.16–1.54)	20 (23.8%)	15 (17.9%)
mri_reface	0/154	0.08 ± 0.04	0.03–0.32 (0.08–0.13)	7 (4.5%)	0 (0%)	0/114	0.10 ± 0.22	0.04–2.23 (0.09–0.14)	6 (5.3%)	2 (1.8%)
PyDeface	0/154	0.08 ± 0.05	0.01–0.32 (0.09–0.19)	5 (3.2%)	0 (0%)	0/114	0.07 ± 0.05	0.01–0.29 (0.07–0.17)	8 (7%)	0 (0%)
spm_deface	0/154	0.07 ± 0.05	0.03–0.33 (0.09–0.18)	10 (6.5%)	0 (0%)	2/114	0.18 ± 0.45	0.03–2.89 (0.09–1.20)	12 (10.5%)	7 (6.1%)

SD, standard deviation; IQR, interquartile range
^aFailed processing denotes defacing crashing (the majority of cases) or veganbagel (i.e., CAT12 for SPM12) processing failing on the defaced image volume (n = 2 for mri_deface and n = 2 for spm_deface). Outliers are reported using Grubbs’s test for each individual approach
^bBy counting any RMSE values which were higher than the 75th percentile of the RMSE values of the benchmark result (0.33)

Back to article page