Table 2.
A summary of the comparison between Freesurfer 6.0, ASHS and CAST automated segmentation methods.
| Automated segmentation methods | Time cost to segment a new subject | Feasible to train a new model? | Hardware requirement to train a new model | Required coregistration | Support multiple modalities | Adaptive to different MRI pulse sequence |
|---|---|---|---|---|---|---|
| Freesurfer 6.0 | A few hours up to one day with -itkthreads 8 setting | Not feasible | Not applicable | Affine and nonlinear | Yes | The same model is used because MRI contrast is discarded |
| ASHS | Approximately 30 min on a 8-core machine | Feasible | Single workstation or computer clusters | Affine and nonlinear | Yes | Performs best when trained on images acquired with the same sequence |
| CAST | Less than 1 min | Feasible | Single workstation with a descent GPU card (e.g. Tesla K40c; GPU is not required for segmentation) | Affine | Yes | Performs best when trained on images acquired with the same sequence |