Quick Fire Session
SCMR 22nd Annual Scientific Sessions
Charlene Mauger
PhD student
The University of Auckland
Kathleen Gilbert, PhD, BEng
Research Fellow
University of Auckland
Aaron Lee, MD, PhD
Assistant Director of Health and Imaging Informatics
Queen Mary University of London
Kenneth Fung, MD, BSc
Clinical Research Fellow
Queen Mary University of London
Valentina Carapella, PhD
Postdoctoral researcher
King's College London
Mihir Sanghvi, MD, BSc
Research Fellow
Queen Mary University of London
Nay Aung, MD
Wellcome Trust Clinical Research Fellow
Queen Mary University of London
Stefan Piechnik, PhD
Head of Advanced Cardiovascular Image Processing, Oxford Centre for Clinical Magnetic Resonance Research
University of Oxford
Stefan Neubauer, MD
Director, Oxford Centre for Clinical Magnetic Resonance Research
University of Oxford
Steffen Petersen, MD, PhD, MSc, FSCMR
Professor of Cardiovascular Medicine and Consultant Cardiologist
Queen Mary University of London
Alistair Young, PhD
Professor
The University of Auckland, New Zealand
Avan Suinesiaputra, PhD
Research Fellow
University of Auckland, New Zealand
Background: Atlases of biventricular shape and function can provide detailed quantitative information on morphological variations and their relationships with disease processes. There has been limited information on the variation of geometry and function with known risk factors. We sought to develop a biventricular statistical shape atlas from the UK Biobank, and investigate the changes undergone in the presence of risk factors.
Methods: A biventricular mesh was customized to contours at end-diastole (ED) and end-systole (ES) from 4,329 UK Biobank cases using a diffeomorphic registration algorithm with automatic breath-hold misregistration correction. Figure 1 presents the atlas construction pipeline. Principal component analysis (PCA) was performed on the concatenation of the mesh coordinates at ED and ES. The strength of associations between biventricular shape and nine cardiovascular risk factors were investigated using logistic regression: presence of diabetes, reduced lung function, high cholesterol, high blood pressure, previous heart attack, stroke, angina, alcohol consumption, and smoking. The regression coefficients associated with the PCA scores were then used as weights to calculate a single morphometric index, allowing continuous representation of the relationship between shape and each risk factor. Two statistical models were also evaluated and compared: one using standard mass and volumes (MassVol), including ED and ES volumes from both ventricles and left ventricle (LV) mass, and the other using the first 50 PCA scores (PCA50).
Results: The dominant atlas shape component was associated with variation in heart size, similar to previous studies [1,2]. The second dominant shape captured tricuspid annular plane systolic excursion. The third and fourth components were associated with the height to width ratio and the base orientation. Biventricular shape had stronger associations with each risk factor than traditional mass and volume measures. Table 1 shows the results of the cross-validation logistic regression models for the PCA50 and the MassVol model. In general, PCA50 model showed higher sensitivity and specificity, precision and F1-score and showed significantly higher AUC for previous heart attack, angina, high cholesterol, smoking, diabetes, and high blood pressure. Figure 2 shows changes undergone by participants with high cholesterol derived from logistic regression. It reveals a diminution of the LV and right ventricle (RV) cavity (possibly due to increased statin use), a torsion of the interventricular septum about the RV axis and an enlargement of the RV around the valves.
Conclusion: By using PCA shapes instead of clinical mass/volumes measure, we were able to identify differences associated with cardiovascular risk factors affecting heart shape using a biventricular atlas, achieving better discriminatory power than standard measures of mass and volume alone.