Structure and validation of novel, automated torso reconstruction pipeline.

A: proposed end-to-end automated 3D torso reconstruction pipeline from 2D standard clinical cardiac magnetic resonance scans. The torso contours are first extracted from the images. A statistical shape model (SSM) is fitted to the contours, which is then optimally deformed. a: spin of the cardiac short axis plane around the torso vertical axis, 0: verticality of the cardiac long axis, γ: tilt of the cardiac short axis plane. B: (i) comparison of Dice coefficient between single stage contouring (blue) and 3 stage segmentation, postprocessing and refinement (red). (ii) surface-to-contour distance for an example case between the automatically reconstructed surface and the automatically generated torso contours. (iii) surface-to-surface distance between the torso mesh created using the automated pipeline and the manually annotated contours for the subjects with the smallest (left) and largest (right) electrode error respectively.

Demographic characteristics of the 1476 subjects from the UK Biobank cohort.

Sex differences in QRS duration in healthy and post-MI subjects.

A, B: mean QRS duration (QRSd) for each ECG lead in healthy and post-MI subjects respectively with women (red circles) and men (cyan triangles). C,D: contribution of anatomical parameters and electrophysiology to sex differences in QRSd for healthy and post-MI subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between male and female populations). E, F: mean QRSd for each ECG lead in men and women respectively for healthy (black squares) and post-MI (red diamonds) subjects. G, H: contribution of anatomical parameters and electrophysiology to differences in QRSd between healthy and post-MI subjects for male and female subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between healthy and post-MI populations).

Sex differences in raw versus corrected QT interval.

Contribution of anatomical parameters and electrophysiology to sex differences in raw QT (A) and corrected QT (B, QTc) intervals for healthy and post-MI subjects in lead I (calculated by multiplying the regression coefficient for each factor by its mean difference between male and female populations).

Sex differences in ECG amplitudes in healthy and post-MI subjects.

A, B: mean STj amplitude (measured at QRS offset) for each ECG lead in healthy and post-MI subjects respectively for women (red circles) and men (cyan triangles). C, D: contribution of anatomical parameters and electrophysiology to sex differences in STj amplitude for healthy and post-MI subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between male and female populations). E, F: mean T wave amplitude (TWA) for each ECG lead in men and women respectively for healthy (black squares) and post-MI (red diamonds) subjects. G, H: contribution of anatomical parameters and electrophysiology to differences in TWA between healthy and post-MI subjects for male and female subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between healthy and post-MI populations).

Sex differences in ECG axis angles in healthy and post-MI subjects

A, B: contribution of anatomical parameters and electrophysiology to differences in R and T axis angles between healthy and post-MI subjects for male and female subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between healthy and post-MI populations). C, D: normalised regression coefficients showing the association between the verticality of the cardiac long axis and the R and T axis angles respectively for the healthy and post-MI populations. Bars shown in red represent regression coefficients significantly different from 0, at a significance level of 0.05. E, F: normalised regression coefficients showing the association between torso volume and the R and T axis angles respectively for the healthy and post-MI populations.

Exclusion pipeline for reconstruction dataset.

Both healthy and post-myocardial infarction (MI) subjects were required to have sufficient images across the view subtypes to reconstruct the torso and cardiac geometries. Subjects also were excluded if the ECG was missing or invalid (all healthy subjects had completed ECG). In the healthy dataset, subjects with at least 1 of 82 disease diagnoses were omitted, including circulatory system disorders, renal disease, genitourinary disorders, and endocrine disorders. The most common reason for exclusion was primary hypertension. Post-MI subjects were excluded if the recorded date of the MI event was after that of the imaging visit where the cardiac magnetic resonance images and ECG were taken.

Depiction of the relative coordinate system in which the heart position was measured.

Left: three principle directions on an example torso-cardiac anatomy. Right: the calculation of the relative superior position of the cardiac centre on an example anatomy. Blue circles: ECG electrodes. Green circles: most superior and inferior electrodes. Green cross: heart centre position.

Depiction of the parameters used to describe the cardiac orientation with respect to the torso.

Upper left: cardiac and torso axes, upper right: cardiac spin orientation calculation, lower left: cardiac verticality orientation calculation, lower right: cardiac tilt orientation calculation.

Anatomical biomarkers and their relationship with age and BMI.

A-I Distributions of anatomical biomarkers. Horizontal lines show statistically significant differences between subpopulation means with arrows pointing to the larger mean. Heat map of correlation coefficients for each geometrical parameter with age (J) and BMI (K). Correlations with p>0.05 are shaded grey. CVol: total cavity volume, LVm: left ventricular mass, TVol: torso volume, x: lateral heart centre position, y: posterior position, z: superior position, Vert: verticality of the cardiac long axis. HM: healthy male, HF: healthy female, PM: post-MI male, and PF: post-MI female.

Distributions of ECG biomarkers.

Distributions of ECG biomarkers for healthy and post-MI populations and proportions of subjects with AV block and pathological Q waves. ST amplitude was measured at the J point (QRS offset), X point (QRS offset + RR interval/16) and E point (QRS offset + RR interval/8). Horizontal lines show statistically significant differences between subpopulation means with arrows pointing to the larger mean. Many distributions show features of non-normality, such as skew, heavy tails or bimodality. Post-MI subjects had longer PQ and QT intervals and P wave duration, and a lower T amplitude for both sexes. Post-MI than healthy males also had a slower heart rate, lower P amplitude, longer QRS duration, left-deviated R axis, and longer QTc. Post-MI subjects were substantially more likely to have AV block and pathological Q waves. The T axis showed a higher variability in post-MI individuals for both sexes. AV: atrioventricular, QTc: corrected QT interval.

QRS duration-anatomy relationship.

Normalised regression coefficients with 95% confidence intervals for QRS duration (QRSd) in all ECG leads for healthy (upper) and post-MI (lower) subjects. CVol: total cavity volume, TVol: torso volume, x, y, z: x (medial), y (posterior), z (superior) coordinate of the heart centre relative to the electrodes, Vert: verticality of the cardiac long axis. Bars shown in red represent regression coefficients that are significantly different from 0, at a significance level of 0.05. QRSd was shortened by a more vertical cardiac orientation (less negative β) for at least one sex in a majority of leads. Increasing cavity volume prolonged the QRSd. QRSd was largely unaffected by adjustments in the cardiac centre location. Unlike healthy subjects, for post-MI individuals QRSd was largely unaffected by cardiac orientation. QRSd again usually did not significantly vary with cardiac location.

Difference between healthy and post-MI STj amplitude, separated by sex.

A, B: mean STj amplitudes for each ECG lead in men and women respectively for healthy (black squares) and post-MI (red diamonds). C, D: contribution of anatomical parameters and electrophysiology to differences in STj amplitudes between healthy and post-MI subjects for male and female subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between healthy and post-MI populations).

Sex differences in ST amplitudes for differing measurement points in healthy and post-MI subjects.

A, B: mean STX amplitude (measured at QRS offset + RR interval/16) for each ECG lead in healthy and post-MI subjects respectively with women (red circles) and men (cyan triangles). C, D: contribution of anatomical parameters and electrophysiology to sex differences in STX amplitude for healthy and post-MI subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between male and female populations). E, F: mean STE amplitude (measured at QRS offset + RR interval/8) for each ECG lead in healthy and post-MI subjects respectively with women (red circles) and men (cyan triangles). G, H: contribution of anatomical parameters and electrophysiology to sex differences in STE amplitude for healthy and post-MI subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between male and female populations).

Sex differences in T wave amplitude in healthy and post-MI subjects.

A, B: mean T wave amplitude (TWA) for each ECG lead in healthy and post-MI subjects respectively with women (red circles) and men (cyan triangles). C,D: contribution of anatomical parameters and electrophysiology to sex differences in TWA for healthy and post-MI subjects (calculated by multiplying the regression coefficient for each factor by its mean difference between male and female populations).

STj amplitude-anatomy relationship.

Normalised regression coefficients with 95% confidence intervals for STj amplitude in all ECG leads for healthy (upper) and post-MI (lower) subjects. LVm: left ventricular mass, TVol: torso volume, x, y, z: x (medial), y (posterior), z (superior) coordinate of the heart centre relative to the electrodes, Vert: verticality of the cardiac long axis. Bars shown in red represent regression coefficients that are significantly different from 0, at a significance level of 0.05. Significant positional correlations generally followed the pattern that the closer the heart centre was to the position of the exploring electrode, the higher the STj amplitude. Increases in left ventricular mass were associated with increased STj amplitude in some of the precordial leads.

T wave amplitude-anatomy relationship.

Normalised regression coefficients with 95% confidence intervals for T wave amplitude (TWA) in all ECG leads for healthy (upper) and post-MI (lower) subjects. CVol: total cavity volume, TVol: torso volume, x, y, z: x (medial), y (posterior), z (superior) coordinate of the heart centre relative to the electrodes, Vert: verticality of the cardiac long axis. Bars shown in red represent regression coefficients that are significantly different from 0, at a significance level of 0.05. TWA in many leads increases as the heart is oriented more vertically (less negative β). Significant positional correlations generally followed the pattern that the closer the heart centre was to the position of the exploring electrode, the higher the TWA.

Axis angles-anatomy relationship.

Normalised regression coefficients with 95% confidence intervals for R axis (top) and T axis (bottom), in healthy (left) and post-MI (right) subjects. For healthy but not post-MI subjects, R axis was negatively associated with torso volume. R axis was significantly affected by cardiac orientation, but not T axis. CVol: total cavity volume, TVol: torso volume, x, y, z: x (medial), y (posterior), z (superior) coordinate of the heart centre relative to the electrodes, Vert: verticality of the cardiac long axis. Bars shown in red represent regression coefficients significantly different from 0, at a significance level of 0.05.

Disease and treatment characteristics of 425 post-MI subjects from the UK Biobank cohort.

Variance inflation factors for linear regression models with total cavity volume.

Variance inflation factors for linear regression models with left ventricular mass.

Mean and standard deviation values of anatomical and ECG biomarkers in each subpopulation.

Associations between age and BMI and anatomical biomarkers.