3554

Correction of transmit-field induced signal inhomogeneity in 3D MP-FLAIR at 7T

Jan Ole Pedersen¹, Vincent O. Boer², Oula Puonti², Jaco M. Zwanenburg³, and Esben Thade Petersen^2,4
¹Philips Healthcare, Copenhagen, Denmark, ²Danish Research Centre for Magnetic Resonance, Centre for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital, Hvidovre, Denmark, ³Department of Radiology, University Medical Center Utrecht, Utrcecht, Netherlands, ⁴Section for Magnetic Resonance, DTU Health Tech, Technical University of Denmark, Kgs Lyngby, Denmark

Synopsis

In order to ease radiological assessment of 7T MP-FLAIR images, we have developed a fully automated bias-field correction that reduces apparent signal loss caused by inhomogeneities in the RF transmit field. Using simulations and measurements of the transmit field, the bias-field can be corrected for without relying on the simplifying assumptions used in a typical bias-field correction. The algorithm is expandable to other TSE-based sequences and adds limited additional scan time.

Introduction

3D Magnetization Prepared FLuid-Attenuated Inversion Recovery (MP-FLAIR) is widely used at 7T as a whole-brain lesion contrast sequence. The combined effect of magnetization preparation, excitation and the turbo-spin echo readout makes the sequence highly dependent on the transmit field (B₁⁺), (figure 2). Image quality is challenged by spatial inhomogeneity in the B₁⁺ field, which causes local loss of signal, alteration of contrast, and broadening of the point spread function (PSF), which complicates radiological assessment of the images [1] (figure 3). Assuming acceptable SNR is pertained in the affected areas, bias-field correction can counter-act the apparent signal loss. Bias-field correction algorithms based solely on the input image itself are widely used in e.g., segmentation pre-processing. While these may address inhomogeneous B₁⁺, receive field (B₁^-), and B₀, they rely on assumptions of the bias-field, for example it being spatially slowly varying, or resemblance of the imaged brain to an atlas. This typically makes them ill-liked by radiologists, as they fear introduction of atypical artefacts not easily distinguishable from pathology, unexpected behavior for atypical images, or degradation of pathological signal characteristics. As an alternative, we show that a B₁⁺ bias-field correction can be performed through acquisition of a B₁⁺ map and simulations that predict signal dependency to B₁⁺.

Method

An extended phase graph simulation of an MP-FLAIR sequence (based on [2]. Parameters are outlined in figure 1) was repeated for multiple relative B₁⁺ amplitudes using literature-values for T₁, T₂ [2] and proton density (PD) [3]. Scaling was only applied to non-adiabatic pulses. The TSE flip-angle train, simulated FLAIR intensity, gray/white matter contrast and simulated PSF are depicted in figure 2.
A B₁⁺ map (DREAM, 3.75mm×3.75mm, slice thickness: 3.50mm, preparation angle: 40^o, scan time: 19 sec [4]) was acquired before 5 repetitions of the MP-FLAIR sequence on a Philips Achieva 7T MR scanner (Best, Netherlands) using a 2Tx/32Rx head coil (Nova Medical, Wilmington, MA, U.S). For each repetition the scanner’s RF gain was manually changed to 0.6, 0.8, 1.0, 1.2 and 1.4 of the initial RF gain. All scans were performed with dielectric pads (7TLF, Multiwave Imaging, Marseille, France) and are depicted in figure 3.
A (20mm)³ Gaussian smoothing kernel was applied to the DREAM B₁⁺ map before it was resampled to the voxel space of the MP-FLAIR image. Values below 40% were rounded up to 40% to avoid noise-amplification where the DREAM sequence had low SNR (typically coinciding with areas of low SNR for the MP-FLAIR). The simulated signal dependency for white and grey matter was averaged after correction for difference in PD and adaptions made to mimic behavior of the scanner’s coil-addition algorithm. A B₁⁺ bias-field was finally generated by voxel-wise interpolation of the measured B₁⁺ map to the simulated MP-FLAIR-dependency to B₁⁺.

Results

The high dependency of MP-FLAIR to B₁⁺ and RF gain at 7T is exemplified in figure 3. Low RF gain leads to good image quality in the center of the brain, but image darkening in the peripheral areas. Contrary, a high RF gain leads to image darkening in the center and improvements peripherally.
Figure 4 depicts the pixel-wise FLAIR intensity versus the measured relative B₁⁺ amplitude for the 5 repeated FLAIR images. The simulated signal dependency used for generation of the bias field is likewise shown. Below 60% and above 140%, the ability of the simulation to predict the measured signal’s dependency of B₁⁺ starts to decrease, which we ascribe to bias inherent to the DREAM sequence for very low and high flip angles [2].
Figure 5 depicts an MP-FLAIR image (RF gain: 1.0) before (top) and after (mid) applying the generated bias field. The measured B₁⁺ map is likewise depicted (bottom). Centrally in the brain, where the B₁⁺ map is hyper-intense (blue arrow), less intensity distortion is seen after correction. In more peripheral areas, where B₁⁺ is hypo-intense (red arrows), the apparent signal intensity loss is partially mitigated, but the corrected images still appear somewhat dark. We ascribe this to the rounding of B₁⁺ done to avoid amplifying noise in low-SNR areas.

Discussion and Conclusion

We have developed a B₁⁺ bias-field correction aimed at easing radiological assessment of 7T MP-FLAIR images. While bias-field correction may improve human interpretation, it does not improve SNR, PSF or contrast affiliated with inhomogeneous B₁⁺, and this approach remains inferior to techniques that successfully homogenize B₁⁺. Such techniques are however typically cumbersome, expensive or prolongs scan time significantly, limiting their use outside advanced research protocols. In comparison, if a B₁⁺ map is not already acquired, the suggested approach adds 20 seconds of additional scan time. Bias at low and high flip-angles may be reduced by using other B₁⁺ mapping techniques or by acquisition of multiple DREAM maps [4], which may be advantageous for more inhomogeneous B₁⁺ fields, e.g., if dielectric pads are not used. While the corrected image shows less signal intensity variation, other sources of local signal variation (e.g., B₁^- and B₀ inhomogeneities) remain uncorrected for. While not investigated, we expect similar performance for other TSE-weighted sequences.
The demonstrated implementation is computationally cheap, and can be implemented directly on the scanner console, easing implementation in the data flow downstream of the MR-scanner.

Acknowledgements

The 7T scanner was donated by the Danish Agency for Science, Technology and Innovation grant no. 0601-01370B, and The John and Birthe Meyer Foundation.

References

[1] Trattnig, et al. "Key clinical benefits of neuroimaging at 7 T." Neuroimage 168 (2018): 477-489.

[2] Saranathan et al. "Optimization of magnetization-prepared 3-dimensional fluid attenuated inversion recovery imaging for lesion detection at 7 T." Investigative radiology 49.5 (2014): 290.

[3] Tofts, Paul S. "PD: proton density of tissue water." Quantitative MRI of the Brain (2003): 85-108.

[4] Olsson et al. "Reducing bias in DREAM flip angle mapping in human brain at 7T by multiple preparation flip angles." Magnetic Resonance Imaging 72 (2020): 71-77.

Figures

Figure 1:
Sequence diagram of the MP-FLAIR sequence. Sequence parameters: MP:107ms, INV: 2200ms, TSE echo-train length: 176, echo-spacing: 3.7ms, repetition time: 7140ms, image resolution: (0.7mm)³FOV(mm): 250x250,190, sagital read-out, ~6:30min scan time. Numbers above RF pulses denotes flip-angles in degrees.

Figure 2:
Top left: The applied flip-angle train. The dotted line denotes center of k-space.

Top right: Signal dependency to B₁⁺ for grey matter (GM), white matter (WM), cortical lesions (LesCor), white matter lesions (LesWM). The contrast (difference between GM and WM after PD-correction) is depicted as a punctured line.

Bottom left: Simulated signal evolution through all echoes for 3 relative scalings of nominal B₁⁺ (average of WM and GM after PD-correction).

Bottom right: Simulated PSF for 3 scalings of B₁⁺ in the phase-encoding direction (average of WM and GM after PD-correction).

Figure 3:
5 repetitions of an MP-FLAIR scan with different RF gains (0.6 – 1.4). For low RF gains, only the mid brain (where B₁⁺ is relatively large) experience close to nominal flip-angles. For RF gains higher than 0.8, the mid brain shows as hypo-intense due to flip-angles being larger than the nominal flip-angle. The lower occipital lobe (where B₁⁺ is relatively small), show as hypo-intense for all but the highest RF gain, where it experiences close to nominal flip angles and shows as hyper-intense, due to the surrounding tissue experiencing higher than nominal flip-angles.

Figure 4:
Simulated versus measured dependency of MP-FLAIR image intensity to B₁⁺. Dots: Measured MP-FLAIR intensities as a function of measured DREAM B1+. Colors depict different RF gains. Other sources of signal loss (e.g. B₁^-/B₀, fluid-content) cause the measured intensities to “fill out” the area under the curve. Black line: Simulated FLAIR intensities. Increasing discrepancy between simulation and measurements is seen below 60% and above 140% of B₁⁺, which is ascribed to inherent bias in the measured B₁⁺. Red line: The MP-FLAIR dependency to B₁⁺ used for bias-field correction.

Figure 5:
Axial slices of a FLAIR image before (top), and after (mid) bias-field correction. The MP-FLAIR images are depicted with identical scaling. The intensity scale refers to percentage of B₁⁺ by the measured DREAM map (bottom). While the bias-field correction addresses B₁⁺ induced intensity loss in the MP-FLAIR images, SNR and WM/GM contrast remain uncorrected. Hypo-intensity due to a large B₁⁺ (blue arrow) appears completely accounted for. Hypo-intensities due to low B₁⁺ (exemplified at red arrows) is partially accounted for and will remain appearing dark in low SNR areas.

Proc. Intl. Soc. Mag. Reson. Med. 29 (2021)

3554