Noise reduction during diffusion tensor imaging of infants

Jordaan, Johannes Petrus

Noise reduction during diffusion tensor imaging of infants

Master Thesis

2019

Abstract

Acoustic noise produced during echo planar imaging (EPI) has been known to reach excessive levels. In addition to causing general patient discomfort and anxiety, this level of noise makes the scanning of young children and infants particularly difficult. Infants are typically scanned while sleeping to minimise motion as they cannot ethically be sedated for research purposes. The extreme noise during MRI acquisitions often cause them to wake before the end of the scanning session. This problem is exacerbated by particularly noisy acquisitions, such as the single shot echo planar diffusion tensor imaging sequence. The main aim of this project was to reduce the noise of this particular acquisition specifically for the scanning of newborn infants. Acoustic noise during MRI acquisitions mainly originates from mechanical vibrations in the gradient coil assemblies due to interactions between the rapidly changing currents applied to the coils and the main static field. A transfer function relating the output acoustic noise spectrum to the gradient excitation input spectrum was developed and used to identify resonant peaks which would amplify coinciding gradient waveform harmonics. In addition to resonant peaks, the transfer function showed significant amplification of frequencies above 1 kHz. In this work, noise reduction was achieved by implementing digital low-pass filters to reduce high-frequency harmonics of the standard trapezoidal gradient waveforms, focusing on the EPI readout portion of a diffusion tensor imaging (DTI) sequence. For comparison purposes, an EPI readout using sinusoidal frequency encoding waveforms and a constant phase encoding blip was also implemented. In addition to reducing produced noise, a passive noise reduction enclosure was built from open cell polyurethane foam mounted in a PVC frame to surround the sleeping infant and act as an acoustic insulation box. Lastly, the effectiveness of introducing pink noise from an external source to mask the abrupt changes in scanner noise, was also investigated. The altered k-space trajectories due to the modifications made to the EPI readout gradient waveforms were corrected through a custom one-dimensional regridding procedure applied along the frequency encoding axis in k-space. Noise reduction was measured with an Optimic 1155 optical microphone from Optoacoustics, attached on top of a cylindrical water phantom inside a 16 channel infant head coil in the isocenter along the Z-direction, facing the bore in the rightleft direction (similar to the orientation that the ears of a sleeping infant would be). Signal-to-noise ratio (SNR) and fractional anisotropy (FA) within the corpus callosum (CC) were compared for images acquired using the standard and modified (filtered and sinusoidal readouts) DTI sequences, the latter each for regridding kernel window sizes of 2 and 4, respectively. The acoustic noise spectra of the filtered and sinusoidal EPI sequences demonstrated a significant reduction in EPI harmonics compared to the standard sequence, but very little difference between each other. Without the foam enclosure, the filtered acquisition with filtered crushers reduced peak sound pressure levels (SPL) by 3.4 and 4 dBA for strong and no fat suppression, respectively, and A-weighted equivalent continuous sound levels (LA,eq) by 2.5 and 2.8 dBA, respectively. Adding the foam enclosure increased peak SPL reduction to 4.8 dBA with fat suppression and 7 dBA without. The sinusoidal sequence performing similarly or marginally (no more than 0.5%) worse than the filtered on all outcomes. SNR measurements in the CC were higher for all volumes of the filtered acquisition compared to the standard, while those of the sinusoidal were similar or slightly lower compared to the standard acquisition. FA values in the CC of the sinusoidal and filtered acquisitions did not differ from those of the standard acquisition (pairwise student’s t-test, all p’s >0.2). For the 16 channel head coil, image reconstruction time increased by only 45 seconds for a regridding kernel width W = 2. Filtering gradient waveforms is an effective technique to reduce acoustic noise during DTI without increasing acquisition time, reducing image quality, or altering FA measures. The proposed method has the potential to be generalized to most gradient waveforms across a variety of sequences. With the addition of the passive noise reduction enclosure, the combined noise reduction could greatly reduce infant anxiety and startling, leading to an increase in the number of infants in whom the acquisition protocol is completed. Acoustic noise produced during echo planar imaging (EPI) has been known to reach excessive levels. In addition to causing general patient discomfort and anxiety, this level of noise makes the scanning of young children and infants particularly difficult. Infants are typically scanned while sleeping to minimise motion as they cannot ethically be sedated for research purposes. The extreme noise during MRI acquisitions often cause them to wake before the end of the scanning session. This problem is exacerbated by particularly noisy acquisitions, such as the single shot echo planar diffusion tensor imaging sequence. The main aim of this project was to reduce the noise of this particular acquisition specifically for the scanning of newborn infants. Acoustic noise during MRI acquisitions mainly originates from mechanical vibrations in the gradient coil assemblies due to interactions between the rapidly changing currents applied to the coils and the main static field. A transfer function relating the output acoustic noise spectrum to the gradient excitation input spectrum was developed and used to identify resonant peaks which would amplify coinciding gradient waveform harmonics. In addition to resonant peaks, the transfer function showed significant amplification of frequencies above 1 kHz. In this work, noise reduction was achieved by implementing digital low-pass filters to reduce high-frequency harmonics of the standard trapezoidal gradient waveforms, focusing on the EPI readout portion of a diffusion tensor imaging (DTI) sequence. For comparison purposes, an EPI readout using sinusoidal frequency encoding waveforms and a constant phase encoding blip was also implemented. In addition to reducing produced noise, a passive noise reduction enclosure was built from open cell polyurethane foam mounted in a PVC frame to surround the sleeping infant and act as an acoustic insulation box. Lastly, the effectiveness of introducing pink noise from an external source to mask the abrupt changes in scanner noise, was also investigated. The altered k-space trajectories due to the modifications made to the EPI readout gradient waveforms were corrected through a custom one-dimensional regridding procedure applied along the frequency encoding axis in k-space. Noise reduction was measured with an Optimic 1155 optical microphone from Optoacoustics, attached on top of a cylindrical water phantom inside a 16 channel iv infant head coil in the isocenter along the Z-direction, facing the bore in the rightleft direction (similar to the orientation that the ears of a sleeping infant would be). Signal-to-noise ratio (SNR) and fractional anisotropy (FA) within the corpus callosum (CC) were compared for images acquired using the standard and modified (filtered and sinusoidal readouts) DTI sequences, the latter each for regridding kernel window sizes of 2 and 4, respectively. The acoustic noise spectra of the filtered and sinusoidal EPI sequences demonstrated a significant reduction in EPI harmonics compared to the standard sequence, but very little difference between each other. Without the foam enclosure, the filtered acquisition with filtered crushers reduced peak sound pressure levels (SPL) by 3.4 and 4 dBA for strong and no fat suppression, respectively, and A-weighted equivalent continuous sound levels (LA,eq) by 2.5 and 2.8 dBA, respectively. Adding the foam enclosure increased peak SPL reduction to 4.8 dBA with fat suppression and 7 dBA without. The sinusoidal sequence performing similarly or marginally (no more than 0.5%) worse than the filtered on all outcomes. SNR measurements in the CC were higher for all volumes of the filtered acquisition compared to the standard, while those of the sinusoidal were similar or slightly lower compared to the standard acquisition. FA values in the CC of the sinusoidal and filtered acquisitions did not differ from those of the standard acquisition (pairwise student’s t-test, all p’s >0.2). For the 16 channel head coil, image reconstruction time increased by only 45 seconds for a regridding kernel width W = 2. Filtering gradient waveforms is an effective technique to reduce acoustic noise during DTI without increasing acquisition time, reducing image quality, or altering FA measures. The proposed method has the potential to be generalized to most gradient waveforms across a variety of sequences. With the addition of the passive noise reduction enclosure, the combined noise reduction could greatly reduce infant anxiety and startling, leading to an increase in the number of infants in whom the acquisition protocol is completed.

Keywords

biology

Reference:

Collections

Masters

Full item page