Automatic reorientation of three-dimensional, transaxial myocardial perfusion SPECT images

We developed a completely automatic technique to reorient transaxial images into short-axis (oblique) myocardial perfusion SPECT images.

The algorithm starts by isolating (segmenting) the left ventricle (LV) myocardium using a combination of iterative clusterification and rule-based location/size/shape criteria. The three-dimensional, mid-myocardial LV surface is initially estimated as the locus of the trilinearly interpolated maxima for the count profiles originating from the center of mass of the segmented LV. The final mid-myocardial surface is obtained by iteratively applying this process, incorporating additional constraints of shape and texture and using the nonsegmented, nonthresholded transaxial image to obtain information on hypoperfused areas of the myocardium. It is then fitted to an ellipsoid, of which the major axis is assumed to represent the long axis of the LV, and the three-dimensional image volume is resliced perpendicularly to it.

The algorithm was retrospectively applied to 400 dual-isotope studies (200 rest 201TI, 200 stress 99mTc-sestamibi) from 200 consecutive patients. Segmentation was successful in 394/400 (98.5%) of the patients. The reproducibility of computer-based reorientation was perfect and significantly better than either intraobserver or interobserver reproducibility.

Automatic reorientation offers the potential for consistently faster and more accurate image processing and analysis and is an important step towards totally operator-less management of myocardial perfusion SPECT data.