12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667 |
- """
- Read hippocampal subfield volumes computed by Freesurfer
- https://surfer.nmr.mgh.harvard.edu/fswiki/HippocampalSubfields
- >>> from freesurfer_volume_reader.freesurfer import HippocampalSubfieldsVolumeFile
- >>>
- >>> for volume_file in HippocampalSubfieldsVolumeFile.find('/my/freesurfer/subjects'):
- >>> print(volume_file.read_volumes_mm3())
- >>> print(volume_file.read_volumes_dataframe())
- """
- import os
- import re
- import typing
- import pandas
- import freesurfer_volume_reader
- class HippocampalSubfieldsVolumeFile(freesurfer_volume_reader.VolumeFile):
- # https://surfer.nmr.mgh.harvard.edu/fswiki/HippocampalSubfields
- FILENAME_PATTERN = r'^(?P<h>[lr])h\.hippoSfVolumes' \
- r'(?P<T1>-T1)?(-(?P<analysis_id>.+?))?\.v10.txt$'
- FILENAME_REGEX = re.compile(FILENAME_PATTERN)
- FILENAME_HEMISPHERE_PREFIX_MAP = {'l': 'left', 'r': 'right'}
- def __init__(self, path: str):
- self._absolute_path = os.path.abspath(path)
- subject_dir_path = os.path.dirname(os.path.dirname(self._absolute_path))
- self.subject = os.path.basename(subject_dir_path)
- filename_match = self.FILENAME_REGEX.match(os.path.basename(path))
- assert filename_match, self._absolute_path
- filename_groups = filename_match.groupdict()
- assert filename_groups['T1'] or filename_groups['analysis_id'], self._absolute_path
- self.hemisphere = self.FILENAME_HEMISPHERE_PREFIX_MAP[filename_groups['h']]
- self.t1_input = filename_groups['T1'] is not None
- self.analysis_id = filename_groups['analysis_id']
- @property
- def absolute_path(self):
- return self._absolute_path
- def read_volumes_mm3(self) -> typing.Dict[str, float]:
- subfield_volumes = {}
- with open(self.absolute_path, 'r') as volume_file:
- for line in volume_file.read().rstrip().split('\n'):
- # https://github.com/freesurfer/freesurfer/blob/release_6_0_0/HippoSF/src/segmentSubjectT1T2_autoEstimateAlveusML.m#L8
- # https://github.com/freesurfer/freesurfer/blob/release_6_0_0/HippoSF/src/segmentSubjectT1T2_autoEstimateAlveusML.m#L1946
- subfield_name, subfield_volume_mm3_str = line.split(' ')
- subfield_volumes[subfield_name] = float(subfield_volume_mm3_str)
- return subfield_volumes
- def read_volumes_dataframe(self) -> pandas.DataFrame:
- volumes_frame = pandas.DataFrame([
- {'subfield': s, 'volume_mm^3': v}
- for s, v in self.read_volumes_mm3().items()
- ])
- volumes_frame['subject'] = self.subject
- volumes_frame['hemisphere'] = self.hemisphere
- # volumes_frame['hemisphere'] = volumes_frame['hemisphere'].astype('category')
- volumes_frame['T1_input'] = self.t1_input
- volumes_frame['analysis_id'] = self.analysis_id
- return volumes_frame
|