Evaluating the quality of brainstem ROI registration using structural and diffusion MRI
Poster No:
1840
Submission Type:
Abstract Submission
Authors:
Yi-An Chen1,2, Lars Kasper1,3, Clement Chow3, Alexandre Boutet3,4,5, Andres Lozano3,4,5, Kamil Uludag3,6,7, Andreea Diaconescu2,1, Sriranga Kashyap3
Institutions:
1Department of Psychology, University of Toronto, Toronto, Canada, 2Krembil Centre for Neuroinformatics, CAMH, Toronto, Canada, 3Krembil Brain Institute, University Health Network, Toronto, Canada, 4Division of Neurosurgery, Toronto Western Hospital, Toronto, Canada, 5Joint Department of Medical Imaging, University of Toronto, Toronto, Canada, 6Department of Medical Biophysics, University of Toronto, Toronto, Canada, 7Center for Neuroscience Imaging Research, Institute for Basic Science & Department of Biomedical Engineering, Sungkyunkwan University, Suwon, Republic of Korea
First Author:
Yi-An Chen
Department of Psychology, University of Toronto|Krembil Centre for Neuroinformatics, CAMH
Toronto, Canada|Toronto, Canada
Department of Psychology, University of Toronto|Krembil Centre for Neuroinformatics, CAMH
Toronto, Canada|Toronto, Canada
Co-Author(s):
Lars Kasper
Department of Psychology, University of Toronto|Krembil Brain Institute, University Health Network
Toronto, Canada|Toronto, Canada
Department of Psychology, University of Toronto|Krembil Brain Institute, University Health Network
Toronto, Canada|Toronto, Canada
Alexandre Boutet
Krembil Brain Institute, University Health Network|Division of Neurosurgery, Toronto Western Hospital|Joint Department of Medical Imaging, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada
Krembil Brain Institute, University Health Network|Division of Neurosurgery, Toronto Western Hospital|Joint Department of Medical Imaging, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada
Andres Lozano
Krembil Brain Institute, University Health Network|Division of Neurosurgery, Toronto Western Hospital|Joint Department of Medical Imaging, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada
Krembil Brain Institute, University Health Network|Division of Neurosurgery, Toronto Western Hospital|Joint Department of Medical Imaging, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada
Kamil Uludag
Krembil Brain Institute, University Health Network|Department of Medical Biophysics, University of Toronto|Center for Neuroscience Imaging Research, Institute for Basic Science & Department of Biomedical Engineering, Sungkyunkwan University
Toronto, Canada|Toronto, Canada|Suwon, Republic of Korea
Krembil Brain Institute, University Health Network|Department of Medical Biophysics, University of Toronto|Center for Neuroscience Imaging Research, Institute for Basic Science & Department of Biomedical Engineering, Sungkyunkwan University
Toronto, Canada|Toronto, Canada|Suwon, Republic of Korea
Andreea Diaconescu
Krembil Centre for Neuroinformatics, CAMH|Department of Psychology, University of Toronto
Toronto, Canada|Toronto, Canada
Krembil Centre for Neuroinformatics, CAMH|Department of Psychology, University of Toronto
Toronto, Canada|Toronto, Canada
Introduction:
Accurate transformation of regions-of-interest (ROIs) from standard to individual subject space is vital in MRI analyses, such as structural volumetry and BOLD fMRI, particularly for brainstem studies. The small size and dense arrangement of brainstem ROIs mean that minor registration errors can significantly impact results. Typically, ROIs defined on a T1-weighted (T1w) template are transformed to subject space using the participant's T1w image, but this method may not be effective for the brainstem due to limited tissue contrast in T1w images. In contrast, diffusion-weighted imaging (DWI) has demonstrated potential in revealing brainstem nuclei and tracts, suggesting enhanced transformation accuracy. Our study assesses the precision of brainstem ROI transformation using three methods: (1) anatomical T1w, (2) b0, and (3) fractional anisotropy (FA) maps from DWI. We focused on the red nucleus and dorsal raphe nucleus. The red nucleus's consistent anatomical definition facilitates manual delineation and ground-truth comparison, while the dorsal raphe nucleus's location near the cerebral aqueduct and fourth ventricle is ideal for assessing misregistration errors.
Methods:
We acquired T1w, DWI, and susceptibility-weighted imaging (SWI) data from 10 healthy participants using a Siemens Prisma 3T scanner. The data were preprocessed using ANTs, Freesurfer, FSL, and MRtrix. Standard ROIs of the red nucleus and the dorsal raphe nucleus were taken from the Brainstem Navigator probabilistic atlas (IIT-space).
For each participant, we aligned the IIT T1w template (IIT-T1w) with the subject T1w using whole-brain non-linear registration to generate a primary transform from IIT to subject space. Next, we obtained brainstem masks from Freesurfer and refined them manually. Then, the IIT-T1w to subject T1w registration was refined by three methods: (1) a brainstem-confined, SyN-only registration using cross-correlation cost function. For methods (2)and (3) the same registration algorithm, cost function and brainstem mask was used to warp the IIT-b0 and IIT-FA template to the subject b0 and FA maps, respectively. The three methods yielded three different refined nonlinear transforms, which were then concatenated with IIT-to-subject whole-brain transform to form three spatial transformations of the brainstem ROIs from the IIT to the subject space. All registrations were done using ANTs.
To quantify the transformation accuracy of the red nuclei ROIs, ground-truth ROIs were manually delineated using the b0 image and the SWI in ITK-SNAP. Dice coefficients of the ground-truth ROI and the three transformed ROIs were computed for each participant. For the dorsal raphe nucleus, the misregistration fraction was calculated as a proxy of registration accuracy. The mis-registration fraction is the volume of the ROI overlapping with the cerebral aqueduct and the fourth ventricle divided by the total ROI volume. The cerebral aqueduct-fourth ventricle masks were obtained using FreeSurfer, with manual editing in ITK-SNAP. Further analyses were conducted using Python libraries ( numpy, nibabel and pandas).
For each participant, we aligned the IIT T1w template (IIT-T1w) with the subject T1w using whole-brain non-linear registration to generate a primary transform from IIT to subject space. Next, we obtained brainstem masks from Freesurfer and refined them manually. Then, the IIT-T1w to subject T1w registration was refined by three methods: (1) a brainstem-confined, SyN-only registration using cross-correlation cost function. For methods (2)and (3) the same registration algorithm, cost function and brainstem mask was used to warp the IIT-b0 and IIT-FA template to the subject b0 and FA maps, respectively. The three methods yielded three different refined nonlinear transforms, which were then concatenated with IIT-to-subject whole-brain transform to form three spatial transformations of the brainstem ROIs from the IIT to the subject space. All registrations were done using ANTs.
To quantify the transformation accuracy of the red nuclei ROIs, ground-truth ROIs were manually delineated using the b0 image and the SWI in ITK-SNAP. Dice coefficients of the ground-truth ROI and the three transformed ROIs were computed for each participant. For the dorsal raphe nucleus, the misregistration fraction was calculated as a proxy of registration accuracy. The mis-registration fraction is the volume of the ROI overlapping with the cerebral aqueduct and the fourth ventricle divided by the total ROI volume. The cerebral aqueduct-fourth ventricle masks were obtained using FreeSurfer, with manual editing in ITK-SNAP. Further analyses were conducted using Python libraries ( numpy, nibabel and pandas).
Results:
The average combined volume of the manually-delineated left and right red nucleus is 713.58 ± 93.21 mm3, which falls within the range reported in the literature. The average Dice coefficients of the red nuclei ROIs transformed through the three methods are: (1) T1w-based: 0.72 ± 0.02; (2) b0-based: 0.75 ± 0.02; (3) FA-based: 0.76 ± 0.05. The average misregistration fractions for the dorsal raphe nucleus ROIs transformed through the three methods are: (1) T1-based: 0.28 ± 0.06; (2) b0-based: 0.35 ± 0.13; (3) FA-based: 0.18 ± 0.07.
Conclusions:
The results suggest that diffusion image-based approaches, particularly the FA-based approach, could outperform the conventional T1w-based approach in achieving higher accuracy of brainstem ROI transformation.
Modeling and Analysis Methods:
Image Registration and Computational Anatomy 1
Methods Development
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Subcortical Structures 2
Keywords:
Brainstem
MRI
Spatial Warping
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Provide references using author date format
Colpan, M. E. (2010), ‘Subthalamic and red nucleus volumes in patients with Parkinson’s disease: do they change with disease progression?’, Parkinsonism & related disorders, 16(6), 398-403.
Kolpakwar, S. (2021), ‘Volumetric analysis of subthalamic nucleus and red nucleus in patients of advanced Parkinson’s disease using SWI sequences’, Surgical Neurology International, 27-Jul;12:377.
Pawlak, M. (2023), ‘Red Nucleus Volume Decrease in Healthy Aging (P12-6.003)’, Neurology, Apr;100 (17 Supplement 2) 2417.
Singh K. (2021), ‘Probabilistic Atlas of the Mesencephalic Reticular Formation, Isthmic Reticular Formation, Microcellular Tegmental Nucleus, Ventral Tegmental Area Nucleus Complex, and Caudal-Rostral Linear Raphe Nucleus Complex in Living Humans from 7 Tesla Magnetic Resonance Imaging’, Brain Connectivity, Oct;11(8):613-623.