Freesurfer pipeline

FreesurferClinicalPipeline

__init__

Freesurfer clinical pipeline

Parameters:

Name	Type	Description	Default
subject	object	The subject object.	required
session	object	The session object.	required
output_path	str	The output path for the pipeline.	required
use_which_t1w	str	Use specific T1w file if multiple are available. Defaults to "".	''

FreesurferPipeline

Bases: FreesurferStatsExtractorMixin

__init__

Freesurfer pipeline

Parameters:

Name	Type	Description	Default
subject	object	Subject object	required
session	object	Session object	required
output_path	str	Output path	required
use_which_t1w	str	Use specific T1w file if multiple are available. Defaults to "".	''
recon_all	bool	Whether to run recon-all. Defaults to True.	True
subregion_ha	bool	Whether to segment hippocampus and amygdala subregions. Defaults to False.	False
subregion_thalamus	bool	Whether to segment thalamus subregions. Defaults to False.	False
subregion_brainstem	bool	Whether to segment brainstem subregions. Defaults to False.	False
subregion_hypothalamus	bool	Whether to segment hypothalamus subunits. Defaults to False.	False
fsqc	bool	Whether to run FSQC. Defaults to False.	False
stats2csv	bool	Whether to convert stats to CSV. Defaults to False.	False
extract_from	str	Path to extract results from. Defaults to "".	''

SynthSRPipeline

__init__

SynthSR pipeline

Parameters:

Name	Type	Description	Default
subject	object	The subject object containing BIDS information.	required
session	object	The session object containing BIDS information.	required
output_path	str	The output path for the pipeline results.	required
use_which_t1w	str	Use specific T1w file if multiple are available. Defaults to "".	''

---

A more detailed description:

Freesurfer: Principle Methodological Considerations

Freesurfer Version

The current pipeline uses the Freesurfer 7-dev version (freesurfer-linux-ubuntu22_x86_64-dev-20240909-9ca95c6). Although we prefer using deep learning-based tools implemented in Freesurfer 8 for more accurate and robust processing, the 8 version requires a larger RAM. For our groups with limited computational resources, we opt for the 7-dev version and parallelize the processing of multiple subjects to optimize resource usage.

External Brain Mask

We found that the default brainmask generated by Freesurfer may not be optimal for all datasets. Therefore, we choose to use an external brain mask with mri_synthstrip to replace the brain extraction step. Notably, we take the --no-csf option in mri_synthstrip to avoid misclassifying the CSF as grey matter (see the figure below).

We compare the results using the external brain mask (mri_synthstrip with --no-csf) versus the default Freesurfer brain mask in Freesurfer 7-dev in n = 111 subjects.

Agreement between the two processing pipelines was evaluated using the intraclass correlation coefficient (ICC(3,1)), and Pearson correlation coefficient. Results are displayed in the figures below (cortical measures from 3 parcellations)

DK

DKT

a2009s

qcache

We add the -qcache option to the recon-all command to generate additional surface-based measures.

QC

We use fsqc by Deep-MI lab for Freesurfer QC.

Subfield Segmentation

The subfield segmentation part includes:

• hippocampus and amygdala
• thalamus
• brainstem
• hypothalamus

CSV Outputs

Although Freesurfer provides aparcstats2table and asegstats2table commands to generate CSV files, we implement our own CSV generation function to have more control over the output format. All .stats files are parsed to generate CSV files (also when subregions stats files are available) for cortical and subcortical measures except for *h.curv.stats.

Freesurfer: recon-all-clinical.sh

Some extra preprocessing steps were implemented to achieve output similar to recon-all, such as cortical morphology metrics. For detailed information on these steps, please refer to the source code.