Separate cardiac phase variation from anatomical variation

Determine how much of the variation captured by the eleven-structure cardiac CT statistical shape model is attributable to cardiac phase rather than inter-subject anatomy, and characterize completion performance for cross-phase conditioning such as predicting an end-systolic atrium from an end-diastolic ventricle.

Background

The cohorts contain static, predominantly ungated CT scans without reliable cardiac-phase information. Consequently, the fitted shape distribution combines anatomical differences with unknown within-cycle motion, while the reported correlations are consistent with size-driven variation but do not identify its cause.

The paper does not evaluate cross-phase completion, in which observed and predicted structures correspond to different cardiac phases. Establishing phase-aware models or phase-conditioned completion would be necessary to determine whether the released model can support such use.