Pipeline Walkthrough

Overview

sci_run_pipeline() orchestrates the full SCI imaging processing workflow in 8 sequential steps. Each step is idempotent and the pipeline can be interrupted and resumed at any point.

Pipeline Steps

Step	Name	Tool	Output
1	DICOM metadata extraction	pydicom	metadata CSV
2	Anonymization (optional)	pydicom	anonymized DICOMs + mapping
3	DICOM to NIfTI conversion	dcm2niix	.nii.gz files
4	Spinal cord segmentation	SCT	_seg.nii.gz
5	Lesion segmentation	SCT (SCIsegV2)	_lesion_seg.nii.gz
6	Vertebral labeling	SCT	_labels-disc.nii.gz
7	Parameter extraction	SCT	CSVs with metrics
8	Merge & export	R	combined dataset

Running the Full Pipeline

library(scimagR)

status <- sci_run_pipeline(
  project_dir = "~/projects/my-sci-study",
  steps = 1:8,
  anonymize = FALSE
)

Running Individual Steps

Run only specific steps by setting the steps argument:

# Just convert DICOMs
sci_run_pipeline("~/projects/my-sci-study", steps = 3)

# Run segmentation steps
sci_run_pipeline("~/projects/my-sci-study", steps = 4:5)

The QC Pause

After Step 5 (lesion segmentation), the pipeline pauses for manual QC. This is critical: automated lesion segmentation must be visually verified before extracting quantitative parameters.

During the QC pause:

Open the QC HTML reports in data/qc/
Grade each segmentation in the QC log
Re-run failed segmentations if needed
Continue with steps 6-8

Checking Pipeline Status

sci_pipeline_status("~/projects/my-sci-study")

This returns a tibble showing which steps are complete for each patient and session.

Artifact Grading Strategy

After visual QC, assign artifact grades (0-4) in the imaging registry:

0 (none): Perfect image quality
1 (mild): Minor artifacts, measurements reliable
2 (moderate): Visible artifacts, measurements acceptable
3 (severe): Major artifacts, measurements unreliable
4 (non-evaluable): Cannot be analyzed

Use filter_evaluable() to exclude sessions with grade > 2:

evaluable <- filter_evaluable(registry, max_artifact = 2)

Coverage Matrix

After processing, check your data coverage:

plot_coverage(registry)

This tile plot shows which modalities are available for each patient at each timepoint, with artifact grades overlaid.

Resuming After Interruption

The pipeline is designed to be resilient to interruptions. Simply re-run with the same arguments — existing outputs will be detected and skipped:

# Re-running skips completed steps automatically
sci_run_pipeline("~/projects/my-sci-study", steps = 1:8)

# Force re-processing with overwrite = TRUE
sci_run_pipeline("~/projects/my-sci-study", steps = 4:5, overwrite = TRUE)