Metric for fast automated relative assessment of motion correction methods for dynamic PET imaging

This work presents a metric for rapid assessment of motion correction quality to assist comparison between alternative motion correction methods for dynamic PET imaging with the high resolution research tomograph (HRRT). The designed metric allows automatic selection between motion correction methods without visual inspection and has been tested on simulated and real data. The metric relies on the sum of absolute voxel-by-voxel differences for consecutive frames. Noise in the reconstructed images can make it difficult to correctly distinguish between different motion correction methods and can significantly affect the numerical result of voxel-by-voxel differences for consecutive frames. To reduce the noise component, a low pass filter was applied by using an optimised Gaussian kernel (the optimisation was based on simulated data using a numerical brain phantom). Results from 26 real scans are reported. A newly improved motion correction is shown to perform better than the formerly used method. From the 26 cases considered, the proposed metric favoured use of the new motion correction for 23 cases, and only 3 cases were better under the old motion correction. The method presents a fast and effective way of comparing the relative efficacy of motion correction methods, without any need of a gold standard or reference image.