Preprocessing
AutoHS is a two-step participant-level workflow. For the clinical rationale, asymmetry index formula, and HS threshold theory, see Background and theory.
Pipeline overview
BIDS T1w → Step 1: Segmentation → Step 2: AI-compute → Reports
Step 1 — Segmentation
Default: FreeSurfer
Container:
freesurfer/freesurfer:7.4.1(Docker) orfreesurfer_7.4.1.sif(Apptainer)Command:
recon-all -autorecon1 -autorecon2-volonlyPost-step:
mri_segstats→stats/aseg.statsTypical runtime: 2–4 hours per subject on CPU
Optional: FastSurfer (--fastsurfer)
Container:
deepmi/fastsurfer:latestorfastsurfer_latest.sifCommand:
run_fastsurfer.sh --seg_only --device cpuStats:
stats/aseg+DKT.statsTypical runtime: 30–60 minutes per subject on CPU
Step 2 — AI-compute
Runs in the AutoHS Python environment (native on Apptainer/HPC, Docker image otherwise). This step implements the quantitative core described in Background and theory:
Parse Left-Hippocampus and Right-Hippocampus volumes (mm³) from
aseg.statsoraseg+DKT.statsCompute asymmetry index: \(\mathrm{AI} = (L - R) / (L + R)\)
Apply volume laterality (±0.05) and HS screening thresholds (Brain Communications, in press)
Generate
report.json,summary.txt,report.pdf, and optional overlays
Processing a subject/session pair
For each T1w file matched by BIDS queries:
Copy input to an isolated work directory
Run segmentation with subject id
job_<label>Run AI-compute referencing segmentation output
Publish reports to
output_dir/autohs/sub-*/ses-*/
Enabling and disabling steps
Flag |
Effect |
|---|---|
|
Use FastSurfer instead of FS |
(default) |
FreeSurfer vol-only recon-all |
What is happening?
During FreeSurfer you may see long pauses at Talairach registration, subcortical segmentation, and white-matter segmentation — this is expected.
During FastSurfer, monitor scripts/deep-seg.log under the work directory for CNN
segmentation progress.
References
See Background and theory for the full scientific background and Citation for the publication describing the hippocampal asymmetry index and HS classification thresholds.