E4528. Spatial Properties of Noise in Prostate DWI Acquired With an Endo-Rectal Coil as Critical Information for Clinical Quality Assurance
Authors
Milica Medved;
The University of Chicago
Aritrick Chatterjee;
The University of Chicago
Ambereen Yousuf;
The University of Chicago
Aytekin Oto;
The University of Chicago
Gregory Karczmar;
The University of Chicago
Objective:
Multiparametric MRI (mp-MRI), with diffusion weighted imaging (DWI)/apparent diffusion coefficient (ADC) maps is an essential diagnostic tool for prostate cancer. Efforts to establish quality assurance processes for prostate mp-MRI are necessary and underway. Estimates of electronic noise are needed but difficult due to a priori unknown spatial heterogeneity of noise propagated via SENSE reconstruction. Regions with relatively uniform signal have been used for noise estimation, e.g., the internal obturator muscle, bladder, or regions of low signal. Here, we propagate Gaussian noise through the reconstruction pipeline to produce noise maps corresponding to noise levels observed in clinical images and use them to evaluate suitability of low signal and bladder locations for noise estimation in DWI.
Materials and Methods:
Nineteen subjects underwent mp-MRI of the prostate on a 3-T scanner, with multielement anterior and posterior coils and an endo-rectal coil. The protocol included a hybrid multidimensional MRI series that mapped DWI signal dependence over a 4 x 4 matrix of TE and b values with maximum values of TE = 200 ms and b = 1500 s/mm<sup>2</sup>. Raw data was captured and postprocessed using the ReconFrame patch and MRecon Matlab library (Gyrotools, Zurich, Switzerland). Simulated k-space datasets containing only random complex Gaussian noise were generated. Both this ‘pure noise’ and actual acquired k-space data were processed through the MRecon pipeline to produce noise maps and DW images, and the per-voxel map of noise standard deviation (SD) values was mapped over 1024 simulation iterations. ROIs were defined at the apex, midgland, and base of prostate, as well as in two areas of low signal and in the bladder and ROI averages of SD were measured and normalized to the average SD observed over all prostate ROIs.
Results:
Noise SD maps generally showed localized regions of high noise, typically laterally to the endorectal coil and on one side predominantly. The geometric structure of noise maps was not consistent with g-factor amplification inherent to SENSE reconstruction, where the noise would be suppressed in the proximity of a coil element. Excellent correspondence in spatial distribution of noise is observed between actual reconstructed DW images and the corresponding noise maps, indicating that the spatial heterogeneity in low signal areas in high TE/ high b-value images results predominantly from electronic noise propagation. The mean bladder ROI-derived estimate of noise was 85%, and the mean estimate of noise derived from the low signal areas was only 17% of the average prostate noise SD.
Conclusion:
Regions of low signal are sometimes used for noise estimation in DWI and ADC maps, but this is unreliable due to the spatial heterogeneity and noise amplification inherent to SENSE reconstruction. We observe that regions of low signal correspond to the regions of low noise and are not suitable for noise level estimation. The bladder ROIs provide closer estimates of prostate DWI noise when an endorectal coil is used.
