Multi-slab whole-brain in vivo 0.35 mm human brain at 7 T to validate acceleration & denoising
Poster No:
1920
Submission Type:
Abstract Submission
Authors:
Omer Faruk Gulban1,2, Logan Dowdle3, Desmond Ho Yan Tse4, Saskia Bollmann5, Benedikt A. Poser1, Rainer Goebel1,2, Dimo Ivanov1
Institutions:
1Maastricht University, Maastricht, Netherlands, 2Brain Innovation, Maastricht, Netherlands, 3Center for Magnetic Resonance Research, Minneapolis, MN, 4Scannexus, Maastricht, Netherlands, 5School of Electrical Engineering and Computer Science, The University of Queensland, Brisbane, Queensland
First Author:
Omer Faruk Gulban
Maastricht University|Brain Innovation
Maastricht, Netherlands|Maastricht, Netherlands
Maastricht University|Brain Innovation
Maastricht, Netherlands|Maastricht, Netherlands
Co-Author(s):
Saskia Bollmann
School of Electrical Engineering and Computer Science, The University of Queensland
Brisbane, Queensland
School of Electrical Engineering and Computer Science, The University of Queensland
Brisbane, Queensland
Rainer Goebel
Maastricht University|Brain Innovation
Maastricht, Netherlands|Maastricht, Netherlands
Maastricht University|Brain Innovation
Maastricht, Netherlands|Maastricht, Netherlands
Introduction:
In vivo imaging at mesoscopic scale (0.1-0.5 mm) reveals the intricate details of the brain angioarchitecture [1]. However, there are 3 major constraints to advance the study of human angioarchitecture in vivo: (I) imaging the mesoscopic intracortical vessels in practicable scanning durations, (II) accurate gray matter segmentation, (III) analysis methods to quantify the vascular details. While some work outlined the path [references within 2, 3]; recently we progressed in all 3 aspects by providing a 0.35 mm T2* dataset where the intracortical angioarchitecture details are visible, developing and providing the analyses to start quantifying the mesoscopic intracortical vessels [3]. However, this work only covered a third of the brain while the total scanning lasted 3 hours/subject. Here, we provide two major advancements to increase the brain coverage while decreasing the overall duration:
1. We concatenate partial brain slabs to achieve whole-brain coverage while keeping each scan time around 10 minutes. We "slab-stitch" in the post-processing to have whole-brain images.
2. We apply structure tensor denoising [4] to a single run of 0.35 mm MP2RAGE to achieve SNR similar to averaging 6 runs (used in [6]).
As a result, we show that 0.35 mm iso. (near) whole brain in vivo human brain T2*- and T1-weighted contrasts can be achieved with high SNR within 1.5 hours. In addition, our low undersampling dataset provides a testbed for exploring additional acceleration and denoising methods to further reduce the total scanning time in the future.
1. We concatenate partial brain slabs to achieve whole-brain coverage while keeping each scan time around 10 minutes. We "slab-stitch" in the post-processing to have whole-brain images.
2. We apply structure tensor denoising [4] to a single run of 0.35 mm MP2RAGE to achieve SNR similar to averaging 6 runs (used in [6]).
As a result, we show that 0.35 mm iso. (near) whole brain in vivo human brain T2*- and T1-weighted contrasts can be achieved with high SNR within 1.5 hours. In addition, our low undersampling dataset provides a testbed for exploring additional acceleration and denoising methods to further reduce the total scanning time in the future.
Methods:
Data was acquired as described in [3] using ME GRE [5] at 7 T with pTx [6]. Six bipolar echos at 0.35 mm iso. were obtained using a low undersampling. For the first participant, in 5 sessions totalling 7.5 hours, we piloted:
- Different slab positioning and angulations for our partial brain coverage slabs (see Fig1A).
- Nr. of averages required for high SNR T2*-weighted contrast in our ME GRE images.
- The effect of GRAPPA 3 compared to 2 in ME GRE images.
- The effect of structure tensor denoising compared to multi-run MP2RAGE averaging.
Based on the quality control, we made the following optimizations:
- Instead of 4 ME GRE runs, 2 runs were deemed sufficient for depicting the angioarchitecture.
- Instead of GRAPPA 2, GRAPPA 3 was deemed satisfactory also for angioarchitecture visualization.
- Instead of 6 MP2RAGE runs, 1 run with structure tensor denoising was deemed sufficient for cortical gray matter segmentation.
Combined, these advancements allowed us to acquire "near whole brain" T2* and T1 contrasts at 0.35 mm iso. within 1.5 hours of a single scanning session. With this optimized protocol, two more participants were scanned.
The partial brain slabs were stitched by zero filling a central slab and then performing overlap-mask-based affine registration (in ITKSNAP [7]). This step is semi automatic, requiring careful quality control by the user (takes an hour for stitching 5 slabs). Slab-stitching of MP2RAGE images was done using the same method after the UNI contrast after denoising (using "segmentator_filters" [9]).
- Different slab positioning and angulations for our partial brain coverage slabs (see Fig1A).
- Nr. of averages required for high SNR T2*-weighted contrast in our ME GRE images.
- The effect of GRAPPA 3 compared to 2 in ME GRE images.
- The effect of structure tensor denoising compared to multi-run MP2RAGE averaging.
Based on the quality control, we made the following optimizations:
- Instead of 4 ME GRE runs, 2 runs were deemed sufficient for depicting the angioarchitecture.
- Instead of GRAPPA 2, GRAPPA 3 was deemed satisfactory also for angioarchitecture visualization.
- Instead of 6 MP2RAGE runs, 1 run with structure tensor denoising was deemed sufficient for cortical gray matter segmentation.
Combined, these advancements allowed us to acquire "near whole brain" T2* and T1 contrasts at 0.35 mm iso. within 1.5 hours of a single scanning session. With this optimized protocol, two more participants were scanned.
The partial brain slabs were stitched by zero filling a central slab and then performing overlap-mask-based affine registration (in ITKSNAP [7]). This step is semi automatic, requiring careful quality control by the user (takes an hour for stitching 5 slabs). Slab-stitching of MP2RAGE images was done using the same method after the UNI contrast after denoising (using "segmentator_filters" [9]).
Results:
Fig1B depicts the intricate vascular details easily visible in our T2* weighted images.
Fig2A shows the image quality after post-processing and slab-stitching. Fig2B demonstrates that GRAPPA 3 data is to a very large extent indistinguishable from GRAPPA 2 data. Fig2C illustrates that structure tensor denoising helps greatly to increase the SNR of a single run MP2RAGE image. Although there is some loss of fine details around the vessels, the benefits overcompensate the drawbacks, especially for doing cortical gray matter segmentation.
Fig2A shows the image quality after post-processing and slab-stitching. Fig2B demonstrates that GRAPPA 3 data is to a very large extent indistinguishable from GRAPPA 2 data. Fig2C illustrates that structure tensor denoising helps greatly to increase the SNR of a single run MP2RAGE image. Although there is some loss of fine details around the vessels, the benefits overcompensate the drawbacks, especially for doing cortical gray matter segmentation.
·Figure 1
·Figure 2
Conclusions:
The results demonstrate excellent data quality despite reducing the total scanning time required 5-fold. However, the in-plane undersampling factor of 3 without partial Fourier is still rather conservative since our additional aim is to provide a reference T2* dataset for future studies employing higher acceleration (e.g. NORDIC [9, 10]).
Modeling and Analysis Methods:
Methods Development 1
Novel Imaging Acquisition Methods:
Anatomical MRI 2
Imaging Methods Other
Keywords:
Acquisition
ANGIOGRAPHY
Cortex
Cortical Layers
MR ANGIOGRAPHY
MRI
MRI PHYSICS
STRUCTURAL MRI
Sub-Cortical
Other - Mesoscopic
1|2Indicates the priority used for review
Provide references using author date format
[2] Bollmann, S., Mattern, H., Bernier, M., Robinson, S.D., Park, D., Speck, O., Polimeni, J.R., 2022. Imaging of the pial arterial vasculature of the human brain in vivo using high-resolution 7T time-of-flight angiography. eLife 11, 2021.06.09.447807. https://doi.org/10.7554/eLife.71186
[3] Gulban, O.F., Bollmann, S., Huber, L. (Renzo), Wagstyl, K., Goebel, R., Poser, B.A., Kay, K., Ivanov, D., 2022. Mesoscopic in vivo human T2* dataset acquired using quantitative MRI at 7 Tesla. NeuroImage 264, 119733. https://doi.org/10.1016/j.neuroimage.2022.119733
[4] Weickert, J., 1998. Anisotropic diffusion in image processing. Teubner Stuttgart.
[5] Eckstein, K., Dymerska, B., Bachrata, B., Bogner, W., Poljanc, K., Trattnig, S., Robinson, S.D., 2018. Computationally Efficient Combination of Multi-channel Phase Data From Multi-echo Acquisitions (ASPIRE). Magnetic resonance in medicine 79, 2996–3006. https://doi.org/10.1002/mrm.26963
[6] Tse, D.H.Y., Wiggins, C.J., Ivanov, D., Brenner, D., Hoffmann, J., Mirkes, C., Shajan, G., Scheffler, K., Uludağ, K., Poser, B.A., 2016. Volumetric imaging with homogenised excitation and static field at 9.4 T. Magnetic Resonance Materials in Physics, Biology and Medicine 29, 333–345. https://doi.org/10.1007/s10334-016-0543-6
[7] Yushkevich, P.A., Piven, J., Hazlett, H.C., Smith, R.G., Ho, S., Gee, J.C., Gerig, G., 2006. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. NeuroImage 31, 1116–28. https://doi.org/10.1016/j.neuroimage.2006.01.015
[8] Gulban, O.F., Schneider, M., Marquardt, I., Haast, R.A.M., De Martino, F., 2018. A scalable method to improve gray matter segmentation at ultra high field MRI. PloS one 13, e0198335. https://doi.org/10.1371/journal.pone.0198335
[9] Vizioli, L., Moeller, S., Dowdle, L., Akçakaya, M., De Martino, F., Yacoub, E., Uğurbil, K., 2021. Lowering the thermal noise barrier in functional brain mapping with magnetic resonance imaging. Nature Communications 12. https://doi.org/10.1038/s41467-021-25431-8
[10] Moeller, S. et al. Locally low-rank denoising in transform domains. in (2023).