Outputs¶
HemiSpec outputs should be predictable enough for downstream statistics, machine learning, and manuscript reporting.
Standard workflow layout¶
hemispec workflow and the GUI write a compact user-facing layout. The default final outputs are four voxel maps per subject plus optional tables and validation reports:
<out-dir>/
voxel_maps/
<subject>_ANS.L.nii.gz
<subject>_ANS.R.nii.gz
<subject>_RNS.L.nii.gz
<subject>_RNS.R.nii.gz
tables/
subject_metric_summary.csv
roi_features_bilateral.csv # when ROI export is enabled and an atlas is available
roi_features_bilateral_wide.csv # when ROI export is enabled and an atlas is available
validation/
hemi_classify/ # only when --run-classifier is enabled
trt/ # only when --run-trt is enabled
DGN reconstructions and one-direction metrics are intermediate implementation details. They are removed by default because the stitched actual/reconstructed recon maps are redundant for most users. To debug or run standalone validation on the merged bilateral maps, use --keep-intermediate in CLI or the matching GUI checkbox:
<out-dir>/
intermediate/
recon/
L_to_R/
R_to_L/
direction_metrics/
L_to_R/
R_to_L/
combined_maps/
<subject>_ANS.nii.gz
<subject>_RNS.nii.gz
Lower-level commands keep their own historical contracts. For example, hemispec compute --save-subject-maps still writes subject_maps/<subject>_ANS.nii.gz and subject_maps/<subject>_RNS.nii.gz under its chosen output directory.
Required metadata roadmap¶
Each run should eventually record:
- HemiSpec version.
- Model bundle version.
- Input paths or dataset identifiers.
- Main parameters.
- Output paths.
- Warnings and validation failures.
This full manifest contract is planned; it should not be described as complete until implemented.