make_synthetic_cohort

Note

This page is a reference documentation. It only explains the function signature, and not how to use it. Please refer to the Habitat Guide and Python API guide (v2.0) for usage.

make_synthetic_cohort(n_subjects: int = 4, modalities: Tuple[str, ...] = ('T1', 'T2'), shape: Tuple[int, int, int] = (32, 32, 32), n_subregions: int = 3, rng: int | Generator | SeedSequence = 0, *, realism: Literal['demo', 'legacy'] = 'demo') → Cohort[source]

Build a deterministic imaging cohort entirely in memory.

Each subject receives its own child seed from SeedSequence.spawn so that per-subject noise differs while the cohort remains reproducible for a fixed master seed. The ROI contains n_subregions intensity blobs with distinct modality profiles to support multi-habitat clustering.

With the default realism="demo", volumes look like soft tissue plus an irregular lesion blob (demo-realistic, not clinical anatomy). Pass realism="legacy" for the older flat ROI / zero-background look.

Parameters:

  • n_subjects – Number of subjects to synthesise.

  • modalities – Modality keys attached to every subject.

  • shape – Cubic grid shape (z, y, x).

  • n_subregions – Number of distinguishable subregions inside the ROI.

  • rng – Master seed controlling the entire cohort.

  • realism – "demo" (default) or "legacy".

Returns:

A Cohort with subjects subj001, subj002, …

Examples

>>> from habit.datasets import make_synthetic_cohort
>>> cohort = make_synthetic_cohort(n_subjects=3, rng=42)
>>> len(cohort)
3
>>> cohort.subject_ids
('subj001', 'subj002', 'subj003')
>>> subject = cohort[0]
>>> sorted(subject.images), sorted(subject.masks)
(['T1', 'T2'], ['tumor'])