Improving assessment of congenital heart disease through rapid patient specific modeling

Abstract

Congenital heart disease is the most common birth defect, with an incidence of 75 in every 1000 births. As a result of improved interventions, 90% of people with congenital heart disease now survive to adulthood. They must undergo regular imaging to assess their biventricular (left and right ventricular) function. Analysis of the images is problematic due to the large variety of shapes and complex geometry. In this paper we extend a biventricular modeling method to improve the analysis of MR images from congenital heart disease patients. We used a subdivision surface method to create three customizable exemplars, representing common manifestations of anatomy, and incorporated these as priors into an interactive biventricular customization procedure. The CHD-specific priors were tested on 60 cases representing a variety of congenital heart diseases for which the gold standard manual contours were available. The introduction of multiple priors showed a significant decrease in analysis time while maintaining good correlation between the two methods (R2 >.82).