Automatic assessment of cardiac function from short-axis MRI: procedure and clinical evaluation

Abstract

Cardiac magnetic resonance imaging (MRI) provides a wealth of morphological and physiological information. Automatic extraction of this information is possible by implementing various image processing techniques. However, existing procedures mostly rely on extensive human interaction and are seldom evaluated on a clinical scale. In this study, a nearly automatic process that extracts physiological parameters from cardiac MR images has been both developed and clinically evaluated. Raw images were obtained in the short-axis view and acquired by a gradient-cho (GE) protocol. In images selected to be analyzed, the only manual step required is the indication of a point in the center of the left ventricle (LV). From a set of such images, the process extracts endocardial and epicardial contours and calculates left ventricular volumes, mass and ejection fraction (EF). The process implements novel approaches to image processing techniques such as thresholding and shape extraction and can be adapted to other acquisition protocols. The process has demonstrated a clear potential for accurate extraction of the endocardial contour but a lower one with respect to the epicardial contour as a result of the low contrast between myocardium and some surrounding tissues, generated by the gradient-echo protocol. The ability of the process to asses physiological parameters has been subjected to a systematic clinical evaluation, which compared parameters, derived manually and automatically, in 10 healthy subjects and 10 patients. The evaluation has indicated that although individual volumes and mass were not accurately assessed, the automatic process has shown high potential for assessing the ejection fraction with relatively high accuracy and reliability.