Table 1.

Summary of several commonly used skull-stripping algorithms for T1-weighted images.

Algorithm	Description of method
Brain extraction tool (BET) (4)	Deformable model which expands from an estimated center of gravity until the brain surface is reached, based on intensity-driven estimates of brain vs. non-brain thresholds. Fractional intensity threshold, its vertical gradient, head radius and center of gravity can be adjusted by the user to improve results.
RObust, learning-based Brain EXtraction (ROBEX) (5)	Learning model using combined generative and discriminative models. Fully data-driven; no user-supplied parameters.
AFNI 3dSkullStrip (6)	Modified version of BET, using non-uniformity correction and edge detection to reduce errors. Provides multiple parameters and flags that the user can adjust to improve skull strip.
Brain surface extraction (BSE) (7)	Uses Marr–Hildreth edge detection after anisotropic diffusion filtering to improve boundary contrast. Semi-automated—displays intermediate results to allow for parameter tuning of filter and edge detector
antsBrainExtraction (8) (https://github.com/ANTsX/ANTs)	Completes brain extraction using N4 intensity normalization, a template and probability map. User must determine which template and brain probability maps work best for their data, although sample files are provided on download site.
FreeSurfer (9)	Combination of watershed (intensity based), deformation and atlas-based techniques to identify and extract brain tissue. User can adjust seed point and watershed threshold, if required.

For a more detailed and comprehensive list of skull-stripping techniques, refer to the following review by Kalavathi and Surya Prasath (10).