The UK GenScan study - population-based imaging genetics research using 3D Cardiac Magnetic Resonance

Antonio de Marvao, Timothy J. W. Dawes, Niall Keenan, Wenjia Bai, Wenzhe Shi, Giuliana Durighel, Tamara Diamond, Laura Monje‐Garcia, Stuart A. Cook, Declan P. O’ReganJournal of Cardiovascular Magnetic Resonance 2013 · DOI

Abstract

Cardiac structure and function result from complex interactions between genes, molecular regulators, and environmental factors. Until recently, imaging-genetics studies had limited power as standard methods of cardiac phenotyping rely on global indices of mass and function, providing only macroscopic descriptors. Manual analysis of large cohorts is time-consuming and subject to interobserver variability. Current cardiac magnetic resonance (CMR) cine imaging techniques are limited by poor through-plane resolution. Here we present a high-resolution 3D left ventricular atlas with automated segmentation developed for the UK GenScan study.

Introduction

In the field of imaging genetics, researchers seek to understand how specific genetic variants influence the physical traits (phenotypes) of the heart. However, traditional approaches have been hindered by two major bottlenecks: resolution gaps and coarse metrics. Standard 2D Cardiac Magnetic Resonance (CMR) imaging often suffers from poor "through-plane" resolution, typically featuring 2mm gaps between 8mm slices. This loss of fine 3D structural detail makes it difficult to detect subtle morphological changes.

Furthermore, most clinical studies rely on global indices, such as total ventricular mass or ejection fraction. While useful for general diagnosis, these macroscopic descriptors can mask regional changes caused by genetic factors. The manual segmentation required for large-scale population studies is also labor-intensive and prone to human error. The UK GenScan study addresses these challenges by introducing a high-resolution 3D atlas and an automated pipeline for precise, voxel-wise analysis.

Methodology

The study compared traditional 2D imaging against a novel 3D approach using data from the Genetic Studies of the Heart and Circulation (GenScan) cohort.

The 3D Sequence and Atlas Construction

Researchers utilized a 3D balanced steady-state free precession (b-SSFP) sequence that captures the entire cardiac volume in a single breath-hold. While the raw voxel size was 2 x 2 x 4 mm, it was reconstructed to an isotropic-like 2 x 2 x 2 mm, providing significantly higher spatial resolution than standard 2D slices. Using ITKsnap software, researchers manually labeled voxels to create a high-resolution 3D left ventricular (LV) reference atlas.

Automated Segmentation and Mapping

To process large cohorts efficiently, the team employed multi-atlas simultaneous segmentation and registration. This system automatically maps the 3D structure of a new patient’s heart by comparing it to the reference atlas. A key innovation is the wall thickness mapping: instead of calculating a single average, the system measures the distance between every point on the epicardial (outer) and endocardial (inner) surfaces. This allows for "point-by-point" statistical comparisons across different patient groups.

Comparison of 2D and 3D segmentation. Panels E and F highlight the difference between the "stair-step" artifacts of 2D slicing and the smooth, anatomically accurate 3D reconstruction.

Key Results and Discussion

The study validated the automated 3D method against manual 2D slicing using a dataset of healthy volunteers (n=19), patients with Dilated Cardiomyopathy (DCM, n=5), and patients with Hypertrophic Cardiomyopathy (HCM, n=5).

Accuracy and Efficiency

The automated 3D volumetry proved highly consistent with manual methods. 2D manual volumes averaged 154 mL, while 3D automated volumes averaged 156 mL (p=0.50), indicating no significant discrepancy. The automated method showed a slight bias of approximately 1% greater volume than the manual method but offered a massive increase in processing speed and reproducibility.

Voxel-Wise Power and Clinical Differentiation

The shift from global averages to voxel-wise statistics represents a significant leap in imaging genetics. By co-registering all hearts to a common 3D space, researchers can perform "heat map" style analyses. This is crucial for genetics, where a mutation might only affect the thickness of a specific region, such as the septum or the apex.

3D wall thickness maps demonstrating the ability to differentiate between healthy controls (A), HCM patients with regional thickening (B), and DCM patients with thinned walls (C).

As shown in the 3D maps, the point-by-point comparison successfully identified distinct patterns of wall thickness differences. HCM patients exhibited characteristic thickening, while DCM patients showed significant thinning compared to the control group.

Conclusion

While this pilot study utilized a relatively small cohort (n=29), it demonstrates the feasibility of high-throughput, population-based imaging genetics. The automated 3D pipeline eliminates the "stair-step" artifacts of traditional 2D CMR and provides the statistical resolution necessary to link genetic variants to regional cardiac morphology. Future work will apply this pipeline to larger populations to map the effects of specific genetic variants on heart function, morphology, and strain.

Summary updated

Abstract

Background Cardiac structure and function results from complex interactions between genes, molecular regulators and environmental factors. Until recently, imaging-genetics studies had limited power as standard methods of cardiac phenotyping rely on global indices of mass and function, providing only macroscopic descriptors. Manual analysis of large cohorts is time consuming and subject to interobserver variability. Current cardiac magnetic resonance (CMR) cine imaging techniques are limited by poor through-plane resolution. Here we present a high resolution 3D left ventricular atlas with automated segmentation.
BibTeX
@article{marvao2013uk,
  author = {Antonio de Marvao and Timothy J. W. Dawes and Niall Keenan and Wenjia Bai and Wenzhe Shi and Giuliana Durighel and Tamara Diamond and Laura Monje‐Garcia and Stuart A. Cook and Declan P. O’Regan},
  title = {{The UK GenScan study - population-based imaging genetics research using 3D Cardiac Magnetic Resonance}},
  journal = {Journal of Cardiovascular Magnetic Resonance},
  year = {2013},
  doi = {10.1186/1532-429x-15-s1-e2},
}
Figure