atlas — cohort atlas generation

Schema: AtlasConfig in src/thesis/core/config/validators.py. Drives the cohort-level atlas workflow.

Field	Type	Default	Constraints	Description
presence_value	float	0.10	0.0 ≤ x ≤ 1.0	Threshold used for the prob_threshold calculation. A voxel is counted as “present” when its normalised value exceeds this threshold.
cov_mean_threshold_pct	float	0.01	0.0 ≤ x ≤ 1.0	Global atlas-mean-map percentage used to suppress low-signal voxels when computing CoV. CV is only computed where the mean exceeds cov_mean_threshold_pct × max(atlas_mean).
normalization_method	NormalizationMethod	waytotal	one of waytotal / streamline_density / max / softmax	Method for normalising streamline density volumes before aggregation. The default YAML value is the lowercase string waytotal (enum NormalizationMethod.WAYTOTAL). See below.
tractography_relpath	str	"tractography/probtrackx2"	—	Relative path under each patient directory where per-patient tractography output lives. The workflow expects <patient>/<tractography_relpath>/<run>/warped_streamlines/fdt_paths.nii.gz (+ waytotal). Override to tractography/mrtrix3 for the MRtrix3 backend.
output_subdir	str	"atlas"	—	Subdirectory under the cohort output directory where the statistical atlas (atlas_mean.nii.gz et al.) is written. Override per backend (e.g. atlas_mrtrix3) when running probtrackx2 and mrtrix3 into the same outputs/ tree to avoid clobbering.

normalization_method values

Value	Description
waytotal	FSL ProbTrackX2 — divide by total streamline count (waytotal). Produces fraction-of-total per voxel.
streamline_density	MRtrix3 counterpart of waytotal. Divides by the value in waytotal, which the MRtrix3 workflow writes as the sum of SIFT2 per-streamline weights (when SIFT2 is enabled, matching the SIFT2-weighted TDI numerator) or the raw streamline count when SIFT2 is off.
max	Divide by the maximum voxel value in the volume. Scales to [0, 1].
softmax	Apply softmax (exp(x) / sum(exp(x))). Produces a probability distribution summing to 1.0 over the volume.

Example

atlas:
  presence_value: 0.10
  cov_mean_threshold_pct: 0.01
  normalization_method: waytotal
  tractography_relpath: tractography/probtrackx2

For MRtrix3:

atlas:
  normalization_method: streamline_density
  tractography_relpath: tractography/mrtrix3

Notes

• The atlas workflow is cohort-level — -p/--patient-id and --all are ignored. The workflow scans the configured output base for per-subject outputs.

• Atlas computation uses pure NumPy. The atlas.dask block has been removed entirely.

• For atlas QC outputs (occupancy, cohort core, CV), see atlas_qc.