Detection of patient motion during tomographic myocardial perfusion imaging

Abstract

We compared the effectiveness of four methods for detecting patient motion during tomographic myocardial perfusion imaging: visual inspection of a cine of the raw data, cross-correlation, diverging squares and a new method called two-dimensional fit. The methods were evaluated for their ability to detect the presence of motion, localize the camera angle at which motion occurred and measure the distance of motion. Patient motion was simulated by shifting motion-free images and then masking their periphery so that the field of view did not move on the image matrix. None of the methods detected 3.25 mm of motion with clinically useful accuracies. Visual inspection, cross-correlation and two-dimensional fit most accurately detected axial patient motion (p < 0.05), whereas cross-correlation most accurately detected lateral motion (p < 0.05). For axial motion, cross-correlation and two-dimensional fit most accurately localized the camera angle at which patient motion occurred (p < 0.05). For lateral motion, cross-correlation most accurately localized patient motion (p < 0.05). Two-dimensional fit measured the distance of axial patient motion to +/- 1.1 mm and measured the distance of lateral motion to +/- 8.7 mm. All other methods frequently overestimated or underestimated the distance of motion by > 13 mm. We conclude that cross-correlation adequately screens tomographic myocardial perfusion studies for both axial and lateral patient motion, although visual inspection is adequate for detection of axial motion. Cross-correlation best localizes the camera angle at which the motion occurred. Two-dimensional fit is the only method studied that accurately measures the distance of motion.