Assessment of Singular Value Decomposition (SVD) Based Denoising on Brain Diffusion Weighted Images Acquired on a 0.35T Magnetic Resonance Imaging-Guided Linear Accelerator
Abstract
Purpose
To assess the effectiveness of SVD-based denoising on the brain Diffusion Weighted images acquired from a 0.35T ViewRay Magnetic Resonance Linac (MRL) and enhance the image quality while preserving the diffusion information.
Methods
Brain DWI scans of five subjects were obtained from a 0.35T ViewRay MRL at Dartmouth Hitchcock Medical Center using a multi-slice echo-planar imaging (EPI) sequence (matrix size100x100x21, FOV= 350x350x190 mm3, TR=3200ms, TE=120ms, bandwidth=1352 HZ, flip angle=90o, 6 averages). Diffusion weighting was applied along three directions using b-values of 0,200,300,500 and 800 s/mm2. Apparent diffusion coefficient (ADC) maps were generated in MATLAB Version R2021a by voxel-wise exponential fitting along each diffusion direction and averaging across directions. SVD was applied on DWI images at different b values. Signal-to-Noise Ratio (SNR) and ADC values were calculated before and after applying SVD in different ROI in the brain. The optimized singular value was obtained by analyzing the SNR gain and ADC accuracy at different singular values. Percentage changes in SNR and ADC values were found from the optimized singular value for all five subjects.
Results
SVD denoising resulted a significant improvement in visual image quality and increased SNR across all subjects with consistent ADC measurements. The average percentage change in SNR of the brain DWI after the implementation of SVD at b-value 800 s/mm2 for all five subjects were found to be 5%,6%,33%,36% and 23%. The percentage change in the ADC value was found to be less than 5% for all five subjects.
Conclusion
The application of SVD-based denoising was effective in resulting SNR gain without compromising the ADC accuracy on brain DWI images obtained from a 0.35T ViewRay MRL. Implementing SVD based denoising improves low field DWI image quality sufficiently to support its potential application in MR guided radiotherapy workflows and longitudinal diffusion monitoring during brain radiotherapy.