Comparison of motion correction algorithms for cardiac SPECT.

Abstract

Patient motion remains a significant source of unsatisfactory cardiac SPECT examinations. The extent to which image recovery can be achieved with correction algorithms is unknown. Nine subjects who had completed motion-free redistribution 201Tlcardiac SPECT subsequently underwent simultaneous dual-isotope (201Tl/99mTc) SPECT with a 99mTc cutaneous point source, while the imaging table was subjected to predefined nonreturning y-translation movements. Cardiac reconstructions, marker reconstructions and marker-compressed dynamic images were generated from the raw data after applying the following correction methods: diverging squares, cross-correlation of the cardiac data and cross-correlation of the marker. Marker cross-correlation performed significantly better than all other methods with good-excellent results in all evaluations. This compared with good-excellent results in none of 27 for the raw data, in 13 of 27 for cardiac cross-correlation and in 7 of 27 for diverging squares (p < 10(-5)). The superiority of the marker-based method was confirmed on analysis of bullseye difference maps and quantitation of residual motion in the point-source data. Motion artifacts can accurately be detected and corrected using cross-correlation of an external point-source. Furthermore, this technique provides useful independent information on the degree of image recovery.