Independent Left Ventricular Morphometric Atlases Show Consistent Relationships with Cardiovascular Risk Factors: A UK Biobank Study

Kathleen Gilbert, Wenjia Bai, Charlène Mauger, Pau Medrano−Gracia, Avan Suinesiaputra, Aaron M. Lee, Mihir M. Sanghvi, Nay Aung, Stefan K. Piechnik, Stefan Neubauer, Steffen E. Petersen, Daniel Rueckert, Alistair A. YoungScientific 2019 · DOI

Abstract

Left ventricular (LV) mass and volume are important indicators of clinical and pre-clinical disease processes. However, much of the shape information present in modern imaging examinations is currently ignored. Morphometric atlases enable precise quantification of shape and function, but there has been no objective comparison of different atlases in the same cohort. We compared two independent LV atlases using MRI scans of 4547 UK Biobank participants: (i) a volume atlas derived by automatic non-rigid registration of image volumes to a common template, and (ii) a surface atlas derived from manually drawn epicardial and endocardial surface contours. The strength of associations between atlas principal components and cardiovascular risk factors (smoking, diabetes, high blood pressure, high cholesterol and angina) were quantified with logistic regression models and five-fold cross validation, using area under the ROC curve (AUC) and Akaike Information Criterion (AIC) metrics. Both atlases exhibited similar principal components, showed similar relationships with risk factors, and had stronger associations than a reference model based on LV mass and volume for all risk factors. Morphometric variations associated with each risk factor could be quantified and visualized and were similar between atlases. UK Biobank LV shape atlases are robust to construction method and show stronger relationships with cardiovascular risk factors than mass and volume.

Introduction

In clinical cardiology, the health of the left ventricle (LV) is typically assessed using two primary metrics: mass and volume. While useful, these are blunt instruments that discard the rich geometric information available in modern 3D MRI data. Before overt heart disease develops, the heart often undergoes subtle "remodeling"—changes in sphericity, wall thickness, or local curvature—in response to risk factors like hypertension or diabetes. Morphometric atlases, or statistical shape models, allow for the quantification of these complex changes. However, different methodologies exist for building these atlases, and it has been unclear whether the choice of method impacts clinical conclusions.

Methods and Materials

This study utilized Cardiac Magnetic Resonance (CMR) scans from 4,547 participants in the UK Biobank. The researchers compared two distinct approaches to atlas construction:

  1. Volume Atlas: An automated approach using non-rigid registration to "warp" 3D image volumes to a common template based on pixel intensity.
  2. Surface Atlas: A traditional approach involving expert manual tracing of heart boundaries (epicardial and endocardial surfaces) to which a mathematical mesh is fitted.
The surface atlas approach involves manual contouring of MRI slices to fit a 3D mesh model.

To analyze the high-dimensional data, Principal Component Analysis (PCA) was employed to identify the primary "modes" of shape variation. Logistic regression with five-fold cross-validation then tested the ability of these modes to predict five cardiovascular risk factors: smoking, diabetes, high blood pressure, high cholesterol, and angina.

Key Results

Methodological Robustness

A primary finding was the remarkable consistency between the two atlas methods. Both identified nearly identical primary modes of variation, such as overall size (Mode 1) and sphericity (Mode 2). This suggests that morphometric modeling is a robust tool, largely independent of the specific algorithm used for construction.

Comparison of volume atlas construction and the resulting modes of variation for both atlas methods.

Superior Predictive Power

In every risk factor category, shape-based models outperformed the traditional mass and volume (MassVol) model. For instance, in predicting high blood pressure, the shape atlases achieved an Area Under the ROC Curve (AUC) of approximately 0.77, significantly higher than the 0.68 achieved by the MassVol model ($p < 0.05$). The first 7-20 PCA modes captured the vast majority of predictive power, indicating that heart remodeling follows predictable geometric patterns.

Discussion and Conclusion

The study demonstrates that high-dimensional shape analysis captures clinically relevant information that traditional metrics miss. By visualizing "signature" shapes associated with specific risks—such as apical bulging in diabetes or dilation in angina—clinicians can better understand the remodeling process.

While the study is cross-sectional and relies partly on self-reported data, the results suggest that "shape scores" could serve as powerful early warning systems in longitudinal patient monitoring. Future work may focus on fully automating the surface atlas pipeline using machine learning to bring these advanced diagnostics closer to routine clinical practice.

Summary updated

Abstract

Left ventricular (LV) mass and volume are important indicators of clinical and pre-clinical disease processes. However, much of the shape information present in modern imaging examinations is currently ignored. Morphometric atlases enable precise quantification of shape and function, but there has been no objective comparison of different atlases in the same cohort. We compared two independent LV atlases using MRI scans of 4547 UK Biobank participants: (i) a volume atlas derived by automatic non-rigid registration of image volumes to a common template, and (ii) a surface atlas derived from manually drawn epicardial and endocardial surface contours. The strength of associations between atlas principal components and cardiovascular risk factors (smoking, diabetes, high blood pressure, high cholesterol and angina) were quantified with logistic regression models and five-fold cross validation, using area under the ROC curve (AUC) and Akaike Information Criterion (AIC) metrics. Both atlases exhibited similar principal components, showed similar relationships with risk factors, and had stronger associations (higher AUC and lower AIC) than a reference model based on LV mass and volume, for all risk factors (DeLong p < 0.05). Morphometric variations associated with each risk factor could be quantified and visualized and were similar between atlases. UK Biobank LV shape atlases are robust to construction method and show stronger relationships with cardiovascular risk factors than mass and volume.
BibTeX
@inproceedings{gilbert2019independent,
  author = {Kathleen Gilbert and Wenjia Bai and Charlène Mauger and Pau Medrano−Gracia and Avan Suinesiaputra and Aaron M. Lee and Mihir M. Sanghvi and Nay Aung and Stefan K. Piechnik and Stefan Neubauer and Steffen E. Petersen and Daniel Rueckert and Alistair A. Young},
  title = {{Independent Left Ventricular Morphometric Atlases Show Consistent Relationships with Cardiovascular Risk Factors: A UK Biobank Study}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-018-37916-6},
}
Figure