Fast Fluid Suppressed T1ρ Imaging of the Human Brain
Eberhard Daniel Pracht1 and Tony Stöcker1,2

1German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, 2University of Bonn, Department of Physics and Astronomy, Bonn, Germany

Synopsis

The longitudinal relaxation time in the rotating frame (T) is a non-invasive biomarker, which is sensitive to slow macromolecular interactions. It might potentially be useful to detect pathological changes in neurodegenerative diseases such as Alzheimer’sDisease. However, long scan times and high SAR limit the application of this technique in a routine setting. Furthermore, high CSF signal modulates the image intensity at CSF/tissue boundaries leading to an overestimation of T1ρ in these regions. To overcome these issues we developed a fast and robust approach to assess T in the human brain at high image resolution.

Purpose

To develop a fast and reliable spin-locking method to assess and quantify the longitudinal relaxation time in therotating frame (T) in the human brain at high image resolution (whole brain acquisition, resolution 1.5mm3 isotropic). The implemented sequence and protocol optimizations (fluid suppression, image deblurring, and scan time reduction) lead to a decrease in the overall SNR of the raw MR images. To counteract the SNR loss, image denoising was investigated in order to increase the accuracy and precision of the resulting T data.

Methods

An in-house developed, extended phase graph (EPG) based based variable flip angle-TSE sequence [1] was equipped with a fluid attenuated T preparation, combining previous approaches from Borthakur and Li [2,3] (figure 1). At the beginning of each TR a saturation recovery preparation is applied in order to reset the longitudinal magnetization. At the end of the saturation time TSAT an inversion pulse is applied to perform a fluid suppression.After the inversion time TINV, a B0 and B1 compensated T module is applied [4], followed by the imaging readout. For fat-suppression a water excitation (utilizing binomial pulses) was employed. A centric radial reordering was used to maintain T weighting in the center of k-space. The refocussing angles along the echo trains were calculated to acquire a constant signal shape for point-spread function (PSF) optimization. All experiments were performed in healthy volunteers on a MAGNETOM Skyra scanner (Siemens GmbH, Healthcare Sector,Erlangen, Germany) using a 32 channel head coil using the following parameters: TSAT/TINV/TR = 1000/811/2600 ms, Turbofactor TF = 176, 1.2 mm$$$^3$$$ isotropic resolution (whole brain coverage), 2D-GRAPPA factor 6 (3x2). Sequence parameters (TSAT/TINV/TF) were optimized via simulations (utilizing Bloch equations and the EPG algorithm). In total five T weighted whole-brain volumes were acquired. The spin-lock sampling scheme was based on the Cramer Rao lower bounds theory [5] (spin-lock times TSL = 1, 25, 50, 75, 100 ms). The total scan time was 3:45min. Before curve fitting a motion correction of the individual T scans [6] and image denoising [7] was performed. In order to legitimate the denoising approach simulations on synthetic data were carried out using noise level and distribution similar to the actual experiments.

Results

Figure 2 depicts the curve fitting results for synthetic data after adding Rician noise (left: Ground truth). T maps are shown with (middle) and without (right) application of denoising before the curve fitting. The fitting error is significantly reduced for the denoised data, resulting in both, a higher precision and a higher accuracy of the T values (see figure 3). Figure 4 shows acquired, native T weighted images before and after denoising. The noise level is significantly lower after denoising, while keeping the image contrast and features, as well as the relative signal intensity between the various spin-lock weightings. Additionally, signal from CSF is effectively suppressed in all T images, allowing a more accurate T quantification at tissue/CSF boundaries. Figure 5 depicts a T map after denoising (left) and the corresponding fitting error (right).

Discussion

It was shown that whole-brain, fluid-suppressed T imaging in short scan times is possible. By implementing a fast and SAR efficient imaging sequence the total scan time was shortened significantly. The loss in SNR, induced during the sequence optimization process (magnetization reset, fluid suppression, image deblurring, and scan time reduction), was compensated by the application of image denoising. The described method does not introduce or remove any features in the images. Hence, an increase in accuracy and precision could be achieved in the quantification process. The achieved gain in acquisition speed is essential, as motion during the the acquisition corrupts the T quantification.This is especially important when targeting neurodegenerative diseases, where subject movement during the scans often occurs.

Acknowledgements

No acknowledgement found.

References

[1] Pracht et al. Proc. Intl. Soc. Mag. Reson. Med. 2013; 21, 249

[2] Borthakur et al. Neuroimage. 2008;41,1199-1205

[3] Li et al. Magn Reson Med. 2008;59,298-307

[4] Witschey et al. J Magn Reson. 2007;186,75-85

[5] Jones et al. J Magn Reson Ser B. 1996;113,25-34

[6] Jenkinson et al. Neuroimage. 2012;62,782-790

[7] Manjon et al. J Magn Reson Imaging. 2010;192-203

Figures

Schematic sequence diagram for fluid-suppressed T imaging

Measured T weighted images. Before and after denoising


Simulation: T ground truth (left), fitting result based on noisy data (middle) and after denoising (right)


1D and 2D histograms of the simulated T maps (whole brain). After denoising T accuracy as well as precision is significantly increased

T map and corresponding relative error after denoising the measured data



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
2831