Population-based studies of myocardial hypertrophy: high resolution cardiovascular magnetic resonance atlases improve statistical power

Antonio de Marvao, Timothy J. W. Dawes, Wenzhe Shi, Christopher Minas, Niall Keenan, Tamara Diamond, Giuliana Durighel, Giovanni Montana, Daniel Rueckert, Stuart A. Cook, Declan P. O’ReganJournal of Cardiovascular Magnetic Resonance 2014 · DOI

Abstract

High spatial resolution cardiovascular magnetic resonance (CMR) with automated phenotyping provides greater power for mapping wall thickness than conventional 2D imaging. This approach enables a significant reduction in the sample size required for studies investigating environmental and genetic determinants of left ventricular (LV) wall thickness.

Introduction

Left ventricular (LV) mass and wall thickness are critical indicators of cardiovascular health and are known to be highly heritable. Identifying the specific genetic variants responsible for these traits is a primary goal of large-scale population studies. However, most current research relies on conventional 2D CMR, which typically utilizes thick 8mm slices with 2mm gaps. This low resolution creates a "coarseness" that makes the imaging insensitive to regional variations in wall thickness—the very changes often seen in the early stages of myocardial hypertrophy. Because 2D measurements are less precise, researchers require massive cohorts to achieve sufficient statistical power, increasing the cost and complexity of genetic association studies.

Methodology

To address these limitations, researchers compared standard 2D Cine CMR with a high-resolution 3D approach in a cohort of 138 healthy volunteers (78 female, 60 male; mean age ~40) recruited via the UK 1000 Cardiac Phenomes project.

3D High-Resolution CMR

The 3D sequence was performed in a single breath-hold, achieving a reconstructed voxel size of 1.2 x 1.2 x 2 mm. In contrast, the standard 2D approach yielded voxels of 1.2 x 1.2 x 8 mm. This higher spatial resolution allows for a more detailed capture of the heart's anatomy without the "step artifacts" caused by misalignment between multiple breath-holds in 2D imaging.

The Multi-Atlas Segmentation Technique

The study employed a multi-atlas segmentation technique to analyze the 3D data. Twenty subjects from the cohort were used to create manually labeled cardiac atlases. An automated algorithm, PatchMatch, was then used to find correspondences between cubic "patches" in these atlases and each new target image.

The automated cardiac segmentation method identifies similarities between cubic "patches" in a pool of manually labeled atlases and target images to define heart boundaries.

This voxel-wise comparison across the population allows for the detection of regional signatures of hypertrophy that global averages might miss.

Results

The 3D automated phenotyping method demonstrated superior performance across several metrics compared to the 2D gold standard.

Accuracy and Reliability

The 3D segmentation achieved a Dice coefficient of 0.95, indicating near-perfect overlap with manual labels, compared to 0.93 for the 2D method. The mean error in defining the epicardial boundary was nearly halved, dropping from 2.2 mm in 2D to 1.3 mm in 3D. Furthermore, the 3D method showed excellent test-retest reliability with an Intraclass Correlation Coefficient (ICC) of 0.97.

The Dice coefficient for both epicardium and endocardium improves as the number of atlases increases, plateauing around 12-20 atlases.

Statistical Power and Sample Size

The most significant finding was the gain in statistical efficiency. To detect a 1 mm difference in wall thickness with 90% power, the 3D method required a median of only 56 subjects, whereas the 2D method required 72. In specific regions of the heart, the 3D approach reduced the required sample size by up to 50%.

Discussion

The transition from 2D to high-resolution 3D CMR atlases represents a significant advancement for population-based cardiac research. By capturing regional signatures of hypertrophy and reducing measurement noise, this technique allows for more precise mapping of wall thickness.

The automated nature of the segmentation not only increases speed but also eliminates the variability inherent in manual 2D measurements. While this study focused on healthy volunteers and the left ventricle, the results suggest that high-resolution 3D phenotyping is ready for large-scale "imaging-genetics" association studies, such as those planned for the UK Biobank. This could significantly accelerate the discovery of genetic determinants of heart structure and disease.

Limitations

Future work should evaluate the performance of these atlases in patients with established pathologies, such as hypertrophic cardiomyopathy, where heart morphology may deviate significantly from the healthy templates used here. Additionally, expanding the atlas pool to include a more diverse global population and applying similar techniques to the right ventricle remain important areas for further investigation.

Summary updated

Abstract

High spatial resolution CMR with automated phenotyping provides greater power for mapping wall thickness than conventional 2D imaging and enables a reduction in the sample size required for studies of environmental and genetic determinants of LV wall thickness.
BibTeX
@article{marvao2014population,
  author = {Antonio de Marvao and Timothy J. W. Dawes and Wenzhe Shi and Christopher Minas and Niall Keenan and Tamara Diamond and Giuliana Durighel and Giovanni Montana and Daniel Rueckert and Stuart A. Cook and Declan P. O’Regan},
  title = {{Population-based studies of myocardial hypertrophy: high resolution cardiovascular magnetic resonance atlases improve statistical power}},
  journal = {Journal of Cardiovascular Magnetic Resonance},
  year = {2014},
  doi = {10.1186/1532-429x-16-16},
}
Figure