Automated detection of the left ventricular region of interest by means of the extraction of typical behaviors in cardiac radionuclide angiographies

An automated process for the detection of the left ventricular end diastolic contour is defined. The extraction of a pure left ventricular behavior, obtained by the application of the Gram-Schmidt orthogonalization process on a set of vectors representing the three main typical behaviors present in the scintigraphic sequence, yields the synthesis of a new factorial image particularly matched for the extraction of the left ventricular region of interest (ROI). An edge following technique with conditional dilation applied simultaneously on the two binary images resulting from processing the new left ventricular factorial and the average images with a Laplacian operator gives a robust method for the detection of the left ventricular ROI. >