Abstract
1326
Objectives Calculation of myocardial blood flow (MBF) in dynamic PET data can be affected by patient motion over the period of the acquisition. Small drift or sudden motion can lead to large errors in the calculation of MBF. We show that the effects can be reduced by an automatic motion correction (MC) algorithm applied prior to MBF calculation.
Methods 10 82-Rb perfusion datasets were divided into frames of [12x5sec, 6x10sec, 4x20sec, 4x40sec] and 10 13-NH3 were divided into frames of [9x10sec, 3x30sec, 1x60sec, 1x120sec], where the datasets covered a range of MBF values. Each dataset had 10 different simulated motion added, consisting of a random 2mm translation and 3 degree rotation per frame as well a drift of 20mm in a constant direction over the duration of the frames. The frames in each simulation were corrected using an automatic MC method, propagating backwards from the final frame, comprising registrations between consecutive frames, until the frames no longer contained data that could be reliably registered. The myocardium and blood pool were automatically segmented from the final frame of each simulation and MBF was calculated using the original and the MC frames using in-house software. Despite the motion, the MBF of each simulation should remain the same if MC is successful.
Results Simulations which led to a poor or failed automatic segmentation were removed from the analysis (7%). The calculation of MBF of the simulated datasets showed that the average coefficient of variation of the MBF across the cases was 21.9% without MC and 9.9% with MC. The average standard deviation of the MBF was 0.26ml/g/min without MC and 0.15ml/g/min with MC, where the minimum and maximum values of MBF using MC were 0.49 and 3.85ml/g/min. Across all the simulations 65% of the frames had MC applied, which corresponds to an average of 87% of the duration of each dynamic scan as the frame duration is not constant.
Conclusions Our analysis shows that automatic registration for motion correction improves the consistency of flow calculations for dynamic cardiac PET