Improving left ventricular border recognition using probability surfaces

The authors present a method for automating the generation of search targets used in finding borders of the left ventricle (LV) from digital-subtraction angiography (DSA) images and thus further automating LV-border determination. This method uses a large clinical database of LV borders to generate a probability surface of expected border locations. A composite image of all observed border points is used to create a set of extraction lines that is overlaid on the LV of interest. Border points are found by using a dynamic search on the density matrix formed by extracting pixel values at the points defined by the extraction matrix. User interaction is only needed to define the valve plane, confirm the final border, and make corrections if needed. LV contours then be used to find end-diastole and end-systole volume and ejection fraction