Robust dynamic susceptibility contrast MR perfusion using 4D nonlinear noise filters

Area of Science:

• Medical Imaging
• Neuroimaging
• Biophysics

Background:

• Dynamic susceptibility contrast (DSC) Magnetic Resonance (MR) perfusion imaging is crucial for assessing cerebral blood flow (CBF).
• Acquired DSC-MR data often suffers from noise, potentially compromising the accuracy of CBF estimates.
• Traditional filtering methods may not adequately address the spatiotemporal (4D) nature of DSC-MR data.

Purpose of the Study:

• To evaluate the efficacy of nonlinear 4D filtering techniques in enhancing the robustness of CBF estimates.
• To compare the performance of 4D-Gaussian and 4D-bilateral filters against unfiltered data and linear spatial filters.
• To determine if advanced filtering can mitigate noise in DSC-MR perfusion data.

Main Methods:

• Development of a digital brain perfusion phantom simulating mixed tissue voxels across various signal-to-noise ratios (SNRs).
• Acquisition of DSC-MR data at 3T from 11 acute ischemic stroke patients.
• Generation of cross-calibrated CBF maps using unfiltered, 4D-Gaussian filtered, and 4D-bilateral filtered DSC-MR data.

Main Results:

• The 4D-bilateral filter achieved the lowest CBF root-mean square error (RMSE) in phantom experiments (4.0 mL/min/100 g) compared to no filtering (5.3 mL/min/100 g) and 4D-Gaussian filtering (6.2 mL/min/100 g).
• The 4D-bilateral filter demonstrated superior image quality in both phantom and patient datasets.
• Linear spatial filters were found to be inappropriate, potentially increasing CBF errors.

Conclusions:

• Nonlinear 4D noise filters are optimally suited for the spatiotemporal characteristics of DSC-MR data.
• The 4D-bilateral filter offers a significant improvement in CBF estimation accuracy and image quality.
• These findings support the use of nonlinear 4D filtering for more reliable neuroperfusion analysis.