Brain masking and registration influence QSM quantification
Poster No:
1982
Submission Type:
Abstract Submission
Authors:
Joo-won Kim1, Dimitri Fiani1, Muhammad Haque2, Griselda Barba Villalobos1, Jacqueline Nguyen1, Stephanie Dinh1, Chadi Calarge1, Junqian Xu1
Institutions:
1Baylor College of Medicine, Houston, TX, 2University of Texas, UTHealth Science and Research Center of Houston, Houston, TX
First Author:
Co-Author(s):
Introduction:
Quantitative susceptibility mapping (QSM) provides unique tissue contrast for subcortical nuclei due to its sensitivity to paramagnetic (e.g., iron) sources in the brain. However, the susceptibility map is influenced by various processing steps. Studies have compared different QSM reconstruction algorithms, however, the effects of image registration and masking have received less attention, especially at the tissue-air interface where major sources of extrinsic susceptibility complicate QSM reconstruction. Here, we investigated the effect of brain masking on susceptibilities in the basal ganglia. In addition, we propose an improved image registration approach to optimize template-based region of interest (ROI) transformation for QSM studies.
Methods:
Twenty-four subjects (13 females, 12-50 years old) underwent two MRI sessions at each of two sites within eight days. A 3D multi-echo GRE was acquired; Site 1: 0.75x0.75x1 mm, TR/TE1/TE2/TE3/TE4/TE5=33/6/11.65/16.98/22.31/27.64 ms; Site 2: 0.7x0.7x1 mm, TR/TE1/TE2/TE3/TE4=30.4/7/14/21/28 ms. T1-weighted MPRAGE was acquired.
QSM reconstruction: The GRE phase images were unwrapped using ROMEO [1]. Starting from an atlas (MNI152) template brain mask transformed into the individual space, we applied UK biobank phase reliability (PR) masking, which is based on a local phase variability map [2]. To find an optimal threshold value for the PR mask, we generated PR masks with threshold values falling between 0.7 and 0.97. Using each mask, the background field removal and dipole inversion were performed with QSMbox [3].
Hybrid template: We created two age-specific templates: adolescent (age < 20) and adult groups. We created T1w-QSM hybrid images [4] with two-step registration (rigid body followed by affine) using ANTs [5]. The individual subject's hybrid image was nonlinearly normalized to a template using ANTs, and ROIs for the putamen (Pu), caudate (Cd), and globus pallidus (Gp) were manually defined on the hybrid templates. This template registration consisted of 3 steps: affine, whole brain nonlinear, and basal ganglia (BG)-specific nonlinear registrations. The BG-specific nonlinear registration, with a manually defined mask tightly surrounding the BG, was necessary to achieve improved BG edge alignment. The template ROIs were transformed into individual spaces.
A template orbital sinus mask was manually created and transformed into individual space. The center of mass of this individually masked unwrapped phase image (>90th-percentile image intensity) was calculated as the center of the orbital sinus. To estimate the remaining orbital sinus effects on QSM, we extracted unwrapped phase values along the line from the center of each ROI to the center of the orbital sinus and fitted the extracted values within the PR mask to a mono-exponential function.
QSM reconstruction: The GRE phase images were unwrapped using ROMEO [1]. Starting from an atlas (MNI152) template brain mask transformed into the individual space, we applied UK biobank phase reliability (PR) masking, which is based on a local phase variability map [2]. To find an optimal threshold value for the PR mask, we generated PR masks with threshold values falling between 0.7 and 0.97. Using each mask, the background field removal and dipole inversion were performed with QSMbox [3].
Hybrid template: We created two age-specific templates: adolescent (age < 20) and adult groups. We created T1w-QSM hybrid images [4] with two-step registration (rigid body followed by affine) using ANTs [5]. The individual subject's hybrid image was nonlinearly normalized to a template using ANTs, and ROIs for the putamen (Pu), caudate (Cd), and globus pallidus (Gp) were manually defined on the hybrid templates. This template registration consisted of 3 steps: affine, whole brain nonlinear, and basal ganglia (BG)-specific nonlinear registrations. The BG-specific nonlinear registration, with a manually defined mask tightly surrounding the BG, was necessary to achieve improved BG edge alignment. The template ROIs were transformed into individual spaces.
A template orbital sinus mask was manually created and transformed into individual space. The center of mass of this individually masked unwrapped phase image (>90th-percentile image intensity) was calculated as the center of the orbital sinus. To estimate the remaining orbital sinus effects on QSM, we extracted unwrapped phase values along the line from the center of each ROI to the center of the orbital sinus and fitted the extracted values within the PR mask to a mono-exponential function.
Results:
The combined effect of improved registration of T1w-to-QSM in creating the individual hybrid image and individual hybrid image-to-template significantly improved the accuracy of the ROI transformation, hence minimizing partial volume effects in QSM quantification.
A higher PR mask threshold (i.e., more restrictive brain masking) yielded lower QSM values, and in certain cases, there was a jump in QSM values within 0.05 PR mask threshold differences. (Fig 2A, site 2, between 0.9 and 0.95). Consistently across different ROIs, there were site differences in the effect of PR mask threshold.
The phase profile along the line from the center of an ROI and the center of the orbital sinus can be well-fitted as a single exponential (Fig 2C).
A higher PR mask threshold (i.e., more restrictive brain masking) yielded lower QSM values, and in certain cases, there was a jump in QSM values within 0.05 PR mask threshold differences. (Fig 2A, site 2, between 0.9 and 0.95). Consistently across different ROIs, there were site differences in the effect of PR mask threshold.
The phase profile along the line from the center of an ROI and the center of the orbital sinus can be well-fitted as a single exponential (Fig 2C).
Conclusions:
Because no single "right" choice for the brain mask in QSM reconstruction exists, brain masking threshold should be carefully evaluated in QSM quantification, potentially with reproducibility metrics, taking into account its site-specific nature. The residual effect of brain masking on different ROIs might be approximated by the rate of phase change between the center of the ROI to the edge of the mask at the nearby sinus.
Modeling and Analysis Methods:
Other Methods 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Subcortical Structures 2
Novel Imaging Acquisition Methods:
Imaging Methods Other
Keywords:
MRI
Sub-Cortical
Other - QSM
1|2Indicates the priority used for review
Provide references using author date format
[2] C. Wang et al., “Phenotypic and genetic associations of quantitative magnetic susceptibility in UK Biobank brain imaging,” Nat. Neurosci., vol. 25, no. 6, Art. no. 6, Jun. 2022, doi: 10.1038/s41593-022-01074-w.
[3] J. Acosta-Cabronero, C. Milovic, H. Mattern, C. Tejos, O. Speck, and M. F. Callaghan, “A robust multi-scale approach to quantitative susceptibility mapping,” NeuroImage, vol. 183, pp. 7–24, Dec. 2018, doi: 10.1016/j.neuroimage.2018.07.065.
[4] J. Hanspach et al., “Methods for the computation of templates from quantitative magnetic susceptibility maps (QSM) – toward improved atlas- and voxel-based analyses (VBA),” J. Magn. Reson. Imaging JMRI, vol. 46, no. 5, pp. 1474–1484, Nov. 2017, doi: 10.1002/jmri.25671.
[5] B. B. Avants, N. Tustison, and G. Song, “Advanced normalization tools (ANTS),” Insight J, vol. 2, pp. 1–35, 2009.