Evaluation of Data-based Motion Correction Techniques for High Temporal Resolution Functional PET

Abstract

Functional Positron Emission Tomography (fPET) data offers novel insights into brain energy demands and molecular connectivity. Recent advances in improving temporal resolutions for this imaging technique have opened up new research possibilities. However, lower signal-to-noise ratios (SNR) inherent to short PET frames bring into question whether current realignment approaches still provide appropriate motion correction. Thus, we aimed to evaluate the effectiveness of standard motion correction methods and explore potential improvements for high temporal resolution fPET with 3s frames. We investigated two techniques aimed at improving the SNR to facilitate more accurate realignment of fPET images, in comparison to conventional motion correction: a deep-learning technique based on the application of a conditional generative adversarial network and an exponentially weighted sliding window average. Performance was evaluated by correlating rigid motion parameters between approaches and with simultaneously acquired fMRI data, and by assessing magnitudes of task-induced activation. Our results indicate that neither of the two methods substantially improve mitigation of motion artefacts. Given the increased computational effort of both techniques, we propose that the standard motion correction procedure is adequate for processing high temporal resolution fPET data. Nevertheless, future development of targeted strategies to enhance motion correction may further advance this imaging technique.