Background: Left ventricular size and shape is important for quantifying cardiac remodelling in response to cardiovascular disease. Geometric remodelling indices have been shown to have prognostic value in predicting adverse events in the clinical literature, but these often describe interrelated shape changes. We developed a novel method for deriving orthogonal remodelling components directly from any (moderately independent) set of clinical remodelling indices.
Results: Six clinical remodelling indices (end-diastolic volume index, sphericity, relative wall thickness, ejection fraction, apical conicity and longitudinal shortening) were evaluated using cardiac magnetic resonance images of 300 patients with myocardial infarction, and 1991 asymptomatic subjects, obtained from the Cardiac Atlas Project. Partial least squares (PLS) regression of left ventricular shape models resulted in remodelling components that were optimally associated with each remodelling index. A Gram–Schmidt orthogonalization process, by which remodelling components were successively removed from the shape space in the order of shape variance explained, resulted in a set of orthonormal remodelling components. Remodelling scores could then be calculated which quantify the amount of each remodelling component present in each case. A one-factor PLS regression led to more de-coupling between scores from the different remodelling components across the entire cohort, and zero correlation between clinical indices and subsequent scores.
Conclusions: The PLS orthogonal remodelling components had similar power to describe differences between myocardial infarction patients and asymptomatic subjects as principal component analysis, but were better associated with well-understood clinical indices of cardiac remodelling. The data and analyses are available from www.cardiacatlas.org.