NPZ timeseries schema
NPZ outputs store time-resolved curves, frame indices, and method-specific arrays that support summary phenotypes such as ED/ES volumes, peak strain, time-to-peak strain, aortic flow timing, and atrial pre-contraction rows.
- Modality
- Time-resolved modalities
- Pipeline step
- Feature extraction and time-series export
- Outputs
- ventricle.npz, atrium.npz, aorta.npz, strain_sax.npz, strain_lax.npz, strain_tagged.npz
- Maturity
- Source-audited reference page
Schema principles
- Keys should be readable and unit-aware where practical.
- Arrays should preserve frame order, temporal sampling, and anatomical meaning.
- Summary phenotype pages should link peak/timing values back to the supporting time-series concept.
- Missing NPZ keys can be valid when a row is conditional, but that condition must be documented.
Current time-series families
| Family | Current examples | Source boundary |
|---|---|---|
| Ventricular timing and volume | LV: Volume [mL], RV: Volume [mL], ED/ES frame and ms keys | eval_ventricular_volume.py |
| Atrial timing and volume | LA/RA volume curves, LA: T_pre_a when ECG timing is accepted | eval_atrial_volume.py |
| Aortic flow timing | Aortic Flow: T_ES [frame], Aortic Flow: T_peak_systole [ms] | eval_phase_contrast.py |
| SAX/LAX strain | circumferential, radial, and longitudinal strain arrays | eval_strain_sax.py, eval_strain_lax.py |
| Tagged strain | basal/mid/apical/global tagged strain arrays | eval_strain_tagged.py |
Time-series QC
Time-series outputs need timing context. Frame order, temporal spacing, and phase labels affect peak, derivative, and time-to-peak features.
Source audit
- NPZ families were checked against
docs/data/output_column_inventory.ymland source-audited modality pages. - Timing-key claims were checked against the relevant feature extraction sources.
- Textbook context boundary: broad clinical textbook context is not surfaced here because this page documents time-series contracts.