Segmentation

Segmentation identifies the masks and labels that CardiacNexus feature extraction treats as measurement inputs. Output filenames and label semantics are part of the phenotype contract.

Modality
All segmented modalities
Pipeline step
Step 2 segmentation
Outputs
Segmentation masks, preprocessed LVOT/phase-contrast images, modality-specific QC inputs
Maturity
Source-audited method page

Current implementation boundary

step2_segment.py generates visit-specific Slurm scripts and only writes segmentation commands when required source images are present and expected segmentation files are missing, unless overwrite is requested. The step covers aortic scout, long-axis 2ch/4ch and 4chamber segmentation, short-axis segmentation, aortic distensibility segmentation, LVOT preprocessing/segmentation, phase-contrast preprocessing/segmentation, and native T1 segmentation.

Tagged MRI is handled separately during tagged feature extraction; it is not segmented through the same step-2 mask-generation path.

Expected masks by modality

ModalityExpected mask or prepared outputDownstream pages
Scout/localizerseg_aortic_scout.nii.gzAortic structure
Long-axis cineseg_la_2ch.nii.gz, seg_la_4ch.nii.gz, seg4_la_4ch.nii.gzAtrial, cross-chamber, annular, LAX strain
Short-axis cineseg_sa.nii.gzVentricular structure/function, mass, wall thickness, SAX strain
Aortic distensibilityseg_aortic_dist.nii.gzAortic stiffness/distensibility
LVOT cinelvot_processed.nii.gz, seg_lvot.nii.gzLVOT aortic-root diameters
Phase-contrast flowpreprocessed flow files and seg_aortic_flow.nii.gzAortic flow and flow displacement
Native T1 mappingseg_shmolli_t1map.nii.gzNative T1 tissue characterization

Contract with feature extraction

Changing segmentation filenames, label values, or image-mask geometry without a versioned migration can silently change volumes, areas, strain curves, and aggregate CSV columns. Feature pages therefore document both the current output labels and the segmentation-dependent caveats.

Segmentation-dependent phenotypes

Most CardiacNexus phenotypes are segmentation-dependent measurements. A plausible numeric value can still be unreliable when the upstream mask boundary is wrong.

Source audit

  • Segmentation script generation, expected filenames, overwrite behavior, and modality boundaries were checked against step2_segment.py.
  • Label semantics and QC caveats were cross-checked against current source-audited modality and phenotype pages.
  • Textbook context boundary: broad clinical textbook context is not surfaced here because this is a segmentation implementation boundary page.