NON-GAUSSIAN SIGNAL NOISE IN MAGNETIC RESONANCE ELASTOGRAPHY

Date
2020-05
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Introduction: Magnetic resonance elastography (MRE) is a phase-contrast MRI technology that creates mechanical property maps of the brain in vivo through induced vibration. Signal noise often degrades the quality of reconstructed images and is a limiting factor in developing higher resolution images. While gaussian noise can be filtered during processing, MRE data also contains non-gaussian random noise which is more difficult to correct. Thus far, signal noise has not been thoroughly investigated in MRE. Several different factors have been proposed to contribute to signal noise in MRE including head motion, cardiac cycle, respiration, and table motion. The purpose of this thesis is to quantify different sources of signal noise that degrade the quality of images in MRE. The ability to understand and evaluate sources of noise will allow for the improvement of MRE as a tool for brain health. Methods: To analyze head motion effects, each subject received three separate MRE scans: (1) both motion encoding gradient (MEG) and vibration off, (2) MEG on and vibration off, (3) both MEG and vibration on. A subject received this protocol twice – one scan with the head restricted with foam pillows (Head Restricted MRE) and the second set without the pillows (Traditional MRE) to compare motion noise. To analyze effects of the cardiac cycle, scans were acquired with 8 slices and 250 repetitions. The subject’s head was secured with foam pillows to limit bulk movement and individual wore a pulse oximeter device to record the cardiac cycle while the motion encoding gradient was recorded through an Arduino device. After scanning, the pulse oximeter data was aligned with the motion encoding gradient and noise data to assess the differences in volume of noise depending on where a person was in the cardiac cycle when the MEG was played. Lastly, in order to understand how nongaussian noise impacts the mechanical property estimation procedure, the correlation coefficient was computed for a voxel in relation to the entire image. Plotting the correlation coefficients in-plane and out-of-plane, a line of best fit was found in order to compute the correlation length. This process was performed for the three subject’s data in the head motion MRE experiments where (1) MEG was turned on and Vibration off and (2) when the MEG was turned on and the Vibration was on. Thus, the average correlation length in-plane and out-of-plane could be a metric to determine the extent non-gaussian noise corrupts MRE images. Results: Overall noise was seen to be reduced when a human subject’s head was packed within the head coil during the MRE scan. During the vibrational scan, a larger amount of noise was observed in the y-direction, likely due to the restriction of movement in the x and z directions. Packing reduces the amount of non-gaussian signal noise present and thus decreases the spread of the signal noise, which can likely improve the capabilities of filtering techniques. In the cardiac cycle assessments, the standard deviation and average noise was greatest in the first cardiac bin. These images likely correspond to systole, meaning motion encoding at the peak of systole leads to the most inaccurate image. In the correlation experiments, noise was observed to be more correlated inplane than out-of-plane. Upwards of 8 pixels was computed as the average correlation length in plane. These differences are likely due to the way the images are acquired, such that adjacent voxels in the plane of the image most likely experience similar nonxiv gaussian sources of signal noise, and thus are very related to one-another. Because NLI uses a 10x10 block of pixels to estimate mechanical properties, noise that is highly correlated in-plane may cause errors that current gaussian filtering techniques do not account for. Conclusion: This is the first report to characterize sources of signal noise in MRE. By individually assessing image, physiological, and vibrational noise, it can be seen how non-gaussian sources of noise lead to errors in mechanical property calculations. The use of foam pillows to restrict left-right head movement may be a physical technique to reduce the volume of signal noise and improve the signal-to-noise ratio of data. Identifying that physiological noise closely follows the cardiac cycle demonstrates how harmonic movement outside of induced mechanical vibration can alter the wavepropagation of MRE and thus alter the data used to calculate mechanical properties. Recognizing this is an error that occurs periodically can lead to techniques to filter or remove cardiac pulsation noise. Lastly, calculating correlation length in-plane and outof- plane of noise can assist in researchers developing better techniques to optimize NLI. The ability to identify and characterize the nature of signal noise in MRE will allow for the translation of MRE to be utilized clinically in the brain.
Description
Keywords
biomedical engineering,NON-GAUSSIAN SIGNAL NOISE, MAGNETIC RESONANCE ELASTOGRAPHY
Citation