Correction, Validation, and Characterization of Motion in Resting-State Functional Magnetic Resonance Images of Pediatric PatientsSchabdach, Jenna (2020) Correction, Validation, and Characterization of Motion in Resting-State Functional Magnetic Resonance Images of Pediatric Patients. Doctoral Dissertation, University of Pittsburgh. (Unpublished) This is the latest version of this item.
AbstractThere are many scenarios, for both clinical and research applications, in which we would like to examine a patient's neurodevelopmental status. Generally, neurodevelopmental evaluations can be performed through psychological testing or in-person assessment with a psychologist. However, these approaches are not applicable in all cases, particularly for many pediatric populations. Researchers are beginning to turn to medical imaging approaches for objectively quantifying a patient's neurodevelopmental status. Resting-state functional magnetic resonance images (rs-fMRIs) can be used to study neuronal networks that are active even when a person is not performing a specific task or reacting to particular stimuli. These image sequences are highly sensitive to motion. Techniques have been developed to prevent patients from moving as well as monitor motion during the scan and correct for the patient's movement after the scan. We focus on the first step of retrospective motion correction: volume registration. The purpose of volume registration is to align the contents of all of the image volumes in the image sequence to the contents of a single volume. Traditionally, all image volumes are directly registered to the chosen stationary image volume. However, this approach does not account for significant differences in patient position between the stationary volume and the other volumes in the sequence. We developed a registration framework based on the concept of a directed acyclic graph. We treat the volumes in the sequence as nodes in a graph where pairs of subsequent volumes are connected via directed edges. This perspective allows us to model the relationships between subsequent volumes and account for them during registration. We applied both registration frameworks to a set of simulated images as well as neurological rs-fMRIs from three clinical populations. The clinical populations were preadolescent, neonatal, and fetal subjects who either were healthy or had congenital heart disease (CHD). The original and registered sequences were compared with respect to their local and global motion. The local motion was measured between every pair of image volumes $i$ and $i+1$ in each sequence using the framewise displacement (FD) and the derivative of the root mean square variance of the signal (DVARS). The global motion across each sequence was measured by calculating the similarity between every pair of image volumes in each sequence. The local motion parameters were compared to a pair of gold standard usability thresholds to determine how each registration framework impacted the usability of every image volume. Both the local and global motion parameters were used to determine how many sequences had statistically significant differences in their motion distributions before and after registration. Additionally, the local and global metrics of the original sequences were clustered to determine if a computer could identify groups of subjects based on their motion parameters. The registration frameworks had different effects on each age group of subjects. We found that the neonatal subjects contained the least amount of motion, while the fetal subjects contained the most motion. The DAG-based registration was most effective at reducing motion in the fetal images. Our clustering analysis showed that the different age groups have different global motion parameters, though lifespan-level patterns related to CHD status could not be detected. Share
Details
Available Versions of this Item
MetricsMonthly Views for the past 3 yearsPlum AnalyticsActions (login required)
|
![[feed]](/images/feed-icon-32x32.png){kind=icon}