Evaluation of semi-automated border detection algorithms for the left ventricular endocardium from magnetic resonance images

Left ventricular volume is an important index for a wide range of cardiovascular diseases. Magnetic resonance imaging is an accurate imaging modality for quantitative measurements of left ventricular volume. However, normally this has to be performed manually and relies on the operator´s experience. We developed two semi-automated border detection algorithms to trace the left ventricular endocardial borders and calculate the volume. Comparing with the manual measurement of the left ventricular volumes, the overall bias+standard deviation was 5.9±20.4ml (algorithm 1) and 23.4±6.9ml (algorithm 2, p<;0.05) in end diastole, 6.2±4.6ml (algorithm 1, p<;0.05) and 11.1±8.9ml (algorithm 2, p<;0.05) in end systole and -0.3±22.8ml (algorithm 1) and 13.6±8.4ml (algorithm 2, p<;0.05) in stroke volume.