Sex-specific shape models to disentangle sex effects and age-associated anatomical variation in biventricular cardiac anatomy

Figures

• 

  Figure 1. Overview of the analysis framework. Input data comprised CMR images from 382 volunteers from the SwissHeart study. Long- and short-axis cine images were segmented using nnU-Net models.29 Three-dimensional thorax scans were resampled to cine planes, segmented with the same networks, and used to correct for breath-hold-induced offsets. Segmentation contours were then employed to fit a biventricular mesh at each cardiac frame. The resulting mesh population was used to derive conventional cardiac biomarkers, as well as anatomical sex-agnostic and sex-specific statistical shape models.

• 

  Figure 2. Age-associated differences in end-diastolic ventricular anatomy in males (top row) and females (bottom row), shown as decade-wise group averages relative to the youngest cohort (18–29 years). The colour field represents the signed radial difference between each age-group mean anatomy and the reference anatomy. Positive values indicate an outward spatial offset relative to the reference surface, along the local endocardium-to-epicardium direction, whereas negative values indicate the opposite direction. These values describe between-group anatomical differences and do not represent myocardial motion.

• 

  Figure 3. Explained variance and sex-wise partitioning of sex-agnostic PCA modes. Bars show the percentage of variance explained by each PCA mode for the sex-agnostic, male-specific, and female-specific shape models. Lines show the corresponding cumulative explained variance. Stacked bars indicate the decomposition of sex-agnostic score variance into male within-sex, female within-sex, and between-sex components for each mode.

• 

  Figure 4. Top five leading anatomical modes for the shape models. Top row: sex-agnostic (PCG); middle row: male sex-specific (PCM); bottom row: female-specific model (PCF). Shape variations are shown for ±4 SDs of each anatomical score, with negative scores in light orange and positive scores in light blue.

• 

  Figure 5. Correlations between anatomical scores and clinical biomarkers in sex-specific and sex-agnostic models. Bubble plots show Pearson correlations between anatomical PCA scores and clinical biomarkers for males, left and females, right. Within each panel, scores from the sex-specific model are shown on the left of the dashed vertical line, and scores from the sex-agnostic model are shown on the right. Bubble size indicates correlation magnitude, and colour indicates correlation direction and strength. Positive correlations are shown in blue and negative correlations in red. Only correlation coefficients with absolute magnitude above 0.2 are labelled.

• 

  Figure 6. Comparison of sex-specific and sex-agnostic anatomical statistical shape models across an increasing number of retained PCA modes. Left: cross-validated age prediction error for sex-specific and sex-agnostic models expressed as RMSE. Right: biomarker prediction error for sex-specific and sex-agnostic models expressed as RMSE for LVEDV, LVM, RVEDV, and end-diastolic sphericity Ψ.