Fully automated assessment of left ventricular volumes and mass from cardiac magnetic resonance images

Quantification of left ventricular (LV) size and function from cardiac magnetic resonance (CMR) images requires manual tracing of LV borders on multiple 2D slices, which is subjective, tedious and time-consuming experience. This paper presents a fully automated method for endocardial and epicardial boundaries detection for the assessment of LV volumes, ejection fraction (EF) and mass from CMR images. The segmentation procedure is based on a combined level set approach initialized by an automatically detected point inside the LV cavity. To validate the proposed technique, myocardial boundaries were manually traced on end-diastolic (ED) and end-systolic (ES) frames by an experienced cardiologist. Bland-Altman analysis and linear regression were used to validate LV volumes, EF and mass and similarity metrics were applied to assess the agreement between manually and automatically detected contours. We found minimal biases and narrow limits of agreement for LV volumes, EF and mass; Dice coefficient, Jaccard index and Hausdorff distance evaluated for 2D ED and ES endocardial and epicardial boundaries showed adequate overlapping. The proposed technique allows fast and accurate assessment of LV volumes, EF and mass as a basis for accurate quantification of LV size and function, and myocardial scar from CMR images.