Your First CST Analysis¶
A complete walkthrough: from a single raw DWI acquisition to a PDF metrics report. Allow 30–60 minutes of wall-clock time on a modern workstation.
Prerequisites¶
csttoolinstalled and available on thePATH. See Installation.- A DWI acquisition with whole-brain coverage including the brainstem, with at least 30 gradient directions and
.bval/.bvecsidecars. See Data requirements. - ~5 GB of free disk space for intermediate outputs.
Throughout this tutorial, ~/data/sub-001/dwi/ stands for the directory containing your raw subject; substitute your own path.
Step 0 — Verify the environment¶
doctor confirms that Python, DIPY, ANTs and supporting libraries are present. fetch-data downloads the FMRIB58_FA and Harvard-Oxford atlases used for registration and ROI extraction. The download runs once per machine.
Step 1 — Inspect the data¶
You should see a quality summary covering b-values, gradient direction coverage, voxel size and brain-coverage estimate. Resolve any error-severity findings before continuing; warnings are informational. See the check-dataset reference for the meaning of each line.
Step 2 — Run the full pipeline¶
The simplest invocation chains every stage end-to-end:
csttool run \
--nifti ~/data/sub-001/dwi/sub-001_dwi.nii.gz \
--out ~/derivatives \
--subject-id sub-001 \
--denoise-method patch2self \
--perform-motion-correction \
--bids-out \
--generate-pdf
What each flag does, briefly:
--bids-out— write outputs in BIDS-Derivatives layout under~/derivatives/csttool/sub-001/.--denoise-method patch2self— modern self-supervised denoising (default). Usenlmeansfor acquisitions with <30 volumes.--perform-motion-correction— affine align all volumes to the first b=0.--generate-pdf— render the HTML report to PDF (requires WeasyPrint; see Troubleshooting if this fails).
The run prints stage banners as it progresses: CHECK → IMPORT → PREPROCESS → TRACK → EXTRACT → METRICS.
Step 3 — Read the report¶
Open the PDF:
The report contains:
- Acquisition header — subject ID, scanner field strength, voxel size, b-values, number of directions.
- Hemispheric metrics table — left and right CST: mean FA, MD, RD, AD, streamline count, volume.
- Bilateral comparison — laterality index per metric and a profile-correlation score.
- Tract profiles — line plots of each scalar along the tract from brainstem to motor cortex.
- QC overlays — axial, sagittal and coronal slices with the extracted bundles overlaid on the FA map.
A symmetric subject typically shows |LI| < 0.05 for FA. Larger asymmetries warrant inspection of the QC overlays — most often they reveal a registration glitch rather than a true biological finding.
Interpreting the profiles
The tract profile is resampled to 100 equidistant points along each streamline, so the x-axis spans the bundle from one endpoint to the other. A consistent left/right offset across the whole profile suggests global tract-wide difference; a localised dip points at a specific anatomical level.
Step 4 — Inspect intermediate outputs (optional)¶
If something in the report looks wrong, follow the chain back:
~/derivatives/csttool/sub-001/preproc/dti_preproc.nii.gz— denoised, motion-corrected DWI.~/derivatives/csttool/sub-001/tracking/whole_brain.trk— whole-brain deterministic tractogram.~/derivatives/csttool/sub-001/extract/cst_{left,right}.trk— extracted bilateral CST.~/derivatives/csttool/sub-001/extract/visualizations/registration_qc.png— atlas-to-subject registration overlay.
Load the .trk files in MI-Brain or Trackvis for visual inspection.
What next¶
- Batch processing — run the same pipeline over multiple subjects.
- Troubleshooting — common pitfalls.
- Tractography — what is happening inside the
trackandextractstages.