Analysis Pipeline¶

The analysis phase is driven by DCE_all.sh. It takes preprocessed images and produces Ktrans permeability maps, QC plots, and HTML/PDF reports for each case and the full population.

Inputs¶

These are the outputs of the preprocessing pipeline:

File
`dce/sub-##_ses-##_desc-bfcz_DCE.nii.gz`
`anat/sub-##_ses-##_space-DCEref_T1map.nii`
`anat/sub-##_ses-##_space-DCEref_VFA.nii.gz`
`dce/sub-##_ses-##_desc-AIF_T1map.nii.gz`
`anat/sub-##_ses-##_space-DCEref_desc-brain_mask.nii.gz`

Main Outputs¶

Output File	Description
`sub-##_ses-##_Ktrans.nii`	Ktrans permeability map (min⁻¹)
`sub-##_ses-##_vp.nii`	Plasma volume fraction map
`case_report.html`	Per-case QC report with visualizations
`population_report.html`	Population-level summary and outlier detection

Running Analysis¶

./DCE_all.sh -d /path/to/rawdata [options]

Typical invocation¶

./DCE_all.sh -d /data/rawdata -s -f -C fullPrep

Pipeline Steps¶

Step 1 — Ktrans Mapping¶

Tool: ROCKETSHIP (MATLAB)

Runs the extended Tofts pharmacokinetic model on the preprocessed DCE series using the T1 map and AIF mask. Produces voxel-wise Ktrans and vp maps. See ROCKETSHIP Integration for details.

Step 2 — Gray Matter & CSF Masking¶

Tools: ANTs antsApplyTransforms, FSL fslmaths

Creates GM and CSF masks from the FAST segmentation, registers them to DCE space, and applies them to exclude non-white-matter voxels from the quantitative analysis. This isolates the WM signal relevant to BBB permeability studies.

Step 3 — QC Analysis¶

Scripts: ktrans_analysis.py, ktrans_report.py

Computes slice-by-slice statistics (mean, median, standard deviation) for T1 and Ktrans within white matter and gray matter masks. Also counts zero-valued voxels and flags slices that may indicate registration or fitting failures.

ktrans_report.py then generates a multi-panel slice visualization of the Ktrans map for inclusion in the case report.

Step 4 — Case Report¶

Script: case_report.py

Generates a self-contained case_report.html for each subject/session. The report includes:

Ktrans slice-by-slice visualizations
Registration QC overlays (VFA-on-DCE, MPRAGE-on-DCE)
AIF curve and quality metrics
T1 map statistics
Motion correction plots (if enabled)
Pipeline parameter summary

Reports are rendered from Jinja2 templates (template.html).

Step 5 — Population Report¶

Script: population_report.py

After all cases complete, aggregates results into a single population_report.html. Includes:

Group-level Ktrans and T1 statistics
Outlier detection and flagging
Missing data tracking
Per-subject summary table with links to case reports
Exportable spreadsheet (.xlsx)

Processing runs concurrently across cases using threading with file locking for safe aggregation.

Optional: iNESMA Smoothing (`-S`)¶

When the -S flag is passed, the DCE input is smoothed using iNESMA before Ktrans fitting. iNESMA is a GPU-accelerated non-local means filter (iNESMA_GPU.py) that uses CUDA via Numba. It reduces noise while preserving spatial structure better than Gaussian smoothing.

GPU required

iNESMA requires a CUDA-capable GPU. The Docker image includes CUDA 13 support.

All Options¶

See the Analysis CLI Reference for the full flag list.