Efficient PCA denoising of spatially correlated redundant MRI data

MRI data used for the study: "Henriques, Ianus, Novello, Jovicich, Jespersen, Shemesh. Efficient PCA denoising of spatially correlated redundant MRI data. Imaging Neuroscience (In Press)."

Preclinical scanner data

All animal experiments for the collection of these datasets were preapproved by the institutional and national authorities and carried out according to European Directive 2010/63.

A mouse brain (C57BL/6J) was extracted via transcardial perfusion with 4% Paraformaldehyde (PFA), immersed in 4% PFA solution for 24 h, washed in Phosphate-Buffered Saline (PBS) solution for at least 24 h, and then placed on a 10 mm NMR tube filled with Flourinert (Sigma Aldrich, Lisbon, PT), which was sealed using paraffin film. 

The MRI experiments were performed on a 16.4 T Bruker Aeon Ascend scanner (Bruker, Karlsruhe, Germany), interfaced with an Avance IIIHD console, and equipped with a gradient system capable of producing up to 3000 mT/m in all directions. A constant temperature of 37oC was maintained throughout the experiments using the probe's variable temperature capability. 

Two distinct diffusion-weighted datasets were then acquired using Bruker's standard "Diffusion Tensor Imaging EPI":

• Dataset1 (MB_exp1.nii and its brain mask MB_exp1_mask.nii): For this dataset, we modulated the amount of spatial correlations by acquiring EPI datasets with parameters optimized to mitigate noise spatial correlations, particularly avoiding k-space undersampling acquisition during EPI's gradient ramps and without using partial Fourier, which minimize regridding.
• Dataset2 (MB_exp2.nii and its brain mask MB_exp2_mask.nii): The second dataset was acquired with identical resolution, number of acquisitions, etc., but with large factors inducing spatial correlations, including k-space sampling during gradient ramps (default Bruker's acquisition and reconstruction procedures for acquisition speed) and with a significant phase partial Fourier factor of 6/8 (note for partial Fourier acquisitions, EPI data is reconstructed with zero-padding, according to the default reconstruction procedures by Bruker's pre-clinical reconstruction software Paravision 6.0.1).

All datasets are acquired for the following diffusion-weighted parameters: 30 gradient directions for b-values 1, 2 and 3 ms/μm2 (Δ = 15 ms, δ = 1.5 ms), and 20 consecutive b-value=0 acquisitions - b-values and diffusion gradient directions are saved in files: MB.bval / MB.bvec.

Other acquisition parameters: TR/TE = 3000/50 ms, 9 coronal slices, Field of View = 12×12 mm2, matrix size 80×80, in-plane voxel resolution of 150×150 μm2, slice thickness = 0.7 mm, number of averages = 2, number of segments = 1, double sampling acquisition.

• Gold standard acquisitions for dataset 2 (MB_exp2_20averages.nii): For a gold standard reference, the second dataset was also repeated for 20 averages. Note, since this dataset is aligned to MB_exp2.nii you can use MB_exp2_mask.nii for its brain mask.

For all datasets, Spatial drifts in the image domain were first corrected using a sub-pixel registration technique (Guizar-Sicairos et al., 2008).

Clinical scanner data

Experiments were approved by the Ethical Committee of the University of Trento and the participant signed an informed consent. 

MRI data was a acquired for a healthy control (male, 54 years) using a 3T MAGNETOM PRISMA scanner (Siemens Healthcare, Erlangen, Germany) equipped with a 64-channel head-neck RF receive coil. 

Diffusion MRI data was acquired using a monopolar single diffusion encoding EPI PGSE (Feinberg et al., 2010; Moeller et al., 2010; Xu et al., 2013) along 30 diffusion gradient directions for five non-zero b-values = 1, 2, 3, 4.5 and 6 ms/μm2 (Δ = 39.1 ms, δ = 26.3 ms) and 17 interspersed b-value=0 acquisitions. b-values and diffusion gradient directions are saved in files: HB.bval / HB.bvec. Note, only the masked version of these dataset (HB_masked.nii and its brain mask HB_mask.nii) is provided to guarantee that data privacy standards are met. For noise maps covering all FOV, the noise maps computed as the std of the 5 first repeating unmasked b = 0 acquisitions are provided in file stdS0i.nii.

Other acquisition parameters were the following: TR/TE = 4000/80 ms, 63 axial slices, Field of View = 220×220 mm2, matrix size 110×110, isotropic resolution of 2 mm, 6/8 phase partial Fourier, parallel imaging with GRAPPA 2, simultaneous multi-slice factor 3. All diffusion MRI data was reconstructed using zero-padding, which is the default procedure for data acquired with partial Fourier above 70%.