Fig 1.
Flowchart diagram of the processing pipelines: The BET brain masks and masked brain images are discarded for the BEaST pipeline and replaced by BEaST brain masks and masked images.
Table 1.
Summary of 5variations of SIENA.
Fig 2.
Top four rows: twelve transverse slices through the voxel-by-voxel average of 910 masks generated for each method (from left to right: Vanilla-SIENA, Popescu-SIENA, Cover-SIENA, CBET-SIENA, and BEaST-SIENA), on spectral colormap. The white and black are the areas with complete agreement while colored voxels show variable degree of disagreement shown on the colorbar. BEaST masks overall show the smallest variability compared to other methods. Bottom four rows: The average difference map from scan-rescan dataset for each method. The difference map was created in the native space using SIENA’s flirt registration matrix. Then the difference maps were nonlinearly transformed into the common ICBM space for averaging. The range for all difference maps was set to a constant; note the darker intensity for BEaST, indicating smaller error.
Fig 3.
Scan-rescan comparison of brain extraction masks (Vanilla-BET, Popescu-BET, Cover-BET, CBET, and BEaST) in terms of (a) percent volume difference, (b) absolute percent volume difference, (c) Jaccard similarity index, and (d) mean surface distance. The colored box plot shows extreme values and interquartile range; the red crosses show the outliers with 2.5 SD limit; and the black lines indicate mean and its 95% confidence interval.
Table 2.
Scan-rescan percent brain volume difference, number of failures, and scan-rescan percent brain volume difference.
Table 3.
Mean (SD) annualized percent brain volume change (PBVC), effect size, number of failures, and required sample size per arm for each method and dataset.
Table 4.
Repeat PBVC results and its comparison to original PBVC.