Fig 1.
Flow chart for the inclusion of UK Biobank subjects with paired cMRI and ECG with the respective studies for each cohort.
Table 1.
Summary of the different definitions of cardiac axes investigated.
Fig 2.
Anatomical axes computation methods.
(A) The labelled anatomical regions of the surface meshes as outputted from the automated pipeline. (B) The 5 distinct definitions used to compute the axes, based on PCA’s first component of biventricular point clouds and different combinations of labelled valvular regions. Note that the apex will have a slightly different location for each of the 3 axes definitions, but it has been drawn as a single one for simplicity of the figure.
Fig 3.
Construction of vectorcardiogram for electrical axis definitions.
(A) Median Beat of 12-lead ECG. (B) Orthogonal leads of VCG with the QRS segment in black. (C) 3D QRS loop with 4 definitions of the electrical orientation axis.
Table 2.
Healthy cohort’s demographic and clinical baseline characteristics, N = 27,832. Data is presented as mean ± standard deviation or N (percentage). BSA = Body surface area, LVEF = Left ventricular ejection fraction, LVEDV = left ventricular end diastolic volume.
Fig 4.
Angular distributions of the anatomical-electrical angular differences in 3D and in the three anatomical planes.
Blue arrows show the angle portrayed in the corresponding plot (ACW is negative). The dashed lines correspond to the average separation between the mean anatomical (red arrow) and electrical (orange arrow) axes in each plane. The narrower spread (lower standard deviation) corresponds to a higher spatial consistency, i.e., a more consistent anatomical-electrical relationship across the population. Note that acute angles were used to avoid artificial extremes due to the + 180/-180 boundary.
Fig 5.
Heatmap of the mean geodesic distance between predicted and true electrical axes for all anatomical-electrical pairs.
Fig 6.
Distribution analysis of the anatomical and electrical axis orientation metrics.
(A) Torso diagram with spherical coordinates (,
) for an example cardiac anatomical axis, where
represents the extent to which the base is pointing to anterior or posterior. (B)(C) Contour plots with regression line and Pearson r, showing the anatomical-electrical linear relationship of
and
. (D) Bivariate (
,
) probability density distributions of electrical and anatomical angles, and their differences (normal-fitted marginal distributions and kernel density estimation contours) indicating complex interplay between the two axes. (The anatomical axis presented here is inverted to allow ease of visualisation.).
Fig 7.
Violin plot representations of ,
,
,
divided by BMI and age groups along with their respective sex distributions.
Left panel shows the distributions for the entire population. Increasing BMI is accompanied by an increase in and
. Shifts in angle between sexes are significant at p < 0.001 for all groups except BMI < 18.5.
Fig 8.
Regression coefficients illustrating the relative influence of BMI, age, and sex on cardiac anatomical and electrical axes.
The standardised coefficient, , indicates the change in the angle per standard deviation of the predictor. Absolute values of the coefficients represent the strength of impact each variable has on the angular metrics. Note that coefficient of sex is not standardised by standard deviation and is for Male compared to Female. All multicollinearity and goodness-of-fit results can be found in Tables A–E in S1 Appendix.
Fig 9.
Effect of primary hypertension on the cardiac anatomical and electrical axes.
(A) Kernel density estimates for the comparison of axes angular measures in healthy subjects and subjects with primary hypertension. “x” markers represent the median of each group (B) Box plot comparison of the metrics. All results were significantly different between the two groups at p < 0.001. Healthy N = 27,832; Primary hypertension N = 3,568.
Fig 10.
Multivariable linear regression results comparing healthy and primary hypertension cohorts (Healthy, R2 = 0.31; Primary hypertension, R2 = 0.18).
Standardised coefficients, , with 95% confidence intervals for each predictor, represent strength of impact of each independent variable on
. BSA (Body Surface Area), AS (arterial stiffness), and MAP (mean arterial pressure). All multicollinearity and goodness-of-fit results can be found in S1.4 in S1 Appendix.
Fig 11.
Manhattan plots of the p-values for correlations between the angular metrics of cardiac anatomical-electrical axes and (A) UK BioBank phenotypes, (B) clinical diagnoses, and (C) breakdown of cardiovascular diagnoses associations.
P-values are from a two-sided t-test.