A high definition anatomical brain template of one individual healthy subject
Poster No:
2212
Submission Type:
Abstract Submission
Authors:
Alexandre Cionca1, Yasser Alemán-Gómez2, Elodie Savary1, Céline Provins1, Patric Hagmann2, Oscar Esteban1
Institutions:
1Lausanne University Hospital and University of Lausanne, Lausanne, Vaud, 2Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL), Lausanne, Switzerland
First Author:
Co-Author(s):
Yasser Alemán-Gómez
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland
Patric Hagmann
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland
Introduction:
Brain templates aggregate averaged features across subjects in a normalized stereotactic space and are critical to formalize prior knowledge in neuroimaging analyses [1]. The most widespread templates have been developed by the Montreal Neurological Institute (MNI), yielding a number of templates referred to with the umbrella term of "MNI space" [2]. Recently, some attention shifted towards emphasizing depth rather than width in datasets with few subjects that have many images [3]. Such "dense sampling" protocols, coupled with modern neuroimaging tools, are effective at enhancing image resolution and provide more accurate surfaces [4]. With this in mind, we propose a high-definition template of a single healthy brain built from 70 MR images with multimodal registration.
Methods:
35 T1-weighted (T1w) and 35 T2-weighted (T2w) anatomical brain images of one individual healthy human male (aged 40) were retrieved from the Human Connectome Phantom (HCPh) dataset, an ongoing Stage 1 Registered Report [5]. MRI scans were acquired on a 3T Siemens Magnetom PrismaFit. T1w images were acquired with a magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence with isotropic voxel size of 0.8x0.8x0.8 [mm3], TR=2.2s, TE=2.55ms, FA=8°. T2w images were acquired at the same spatial resolution with a Sampling Perfection with Application optimized Contrast (T2-SPACE), TR=3.2s, TE=413ms.
The images were first corrected for distortion within the internal processing of the scanner (3D distortion correction). Then, correction for intensity non-uniformity and denoising were applied using N4BiasFieldCorrection [6] and DenoiseImage [7] from ANTs. Finally, the intensity of white matter voxels was normalized to values around a value of 110 using mri_normalize from FreeSurfer [8].
The template was estimated in a multivariate (T1w and T2w) approach using antsMultivariateTemplateConstruction2.sh script from ANTs [9]. To first create an initial template, all images are averaged to create a reference and each individual image is then registered to this reference using a rigid body transformation. Then, the initial template is refined in an iterative process where each image is registered to the template using affine transformations to maximize cross-correlation.
An alternative, high-dimension template has been computed by creating a 3D grid of 504x576x384 voxels of size 0.4x0.4x0.4 [mm3] and projecting the center of each voxel to the individual maps for interpolation. The distance between the projected grid coordinates and the nearest image voxel was used as interpolation weight.
The HCPh dataset will be publicly released at the end of Stage 2 of the corresponding registered report. In addition, the template and individual images will be released. The code used for template generation is openly available at https://github.com/acionca/hcph-template.
The images were first corrected for distortion within the internal processing of the scanner (3D distortion correction). Then, correction for intensity non-uniformity and denoising were applied using N4BiasFieldCorrection [6] and DenoiseImage [7] from ANTs. Finally, the intensity of white matter voxels was normalized to values around a value of 110 using mri_normalize from FreeSurfer [8].
The template was estimated in a multivariate (T1w and T2w) approach using antsMultivariateTemplateConstruction2.sh script from ANTs [9]. To first create an initial template, all images are averaged to create a reference and each individual image is then registered to this reference using a rigid body transformation. Then, the initial template is refined in an iterative process where each image is registered to the template using affine transformations to maximize cross-correlation.
An alternative, high-dimension template has been computed by creating a 3D grid of 504x576x384 voxels of size 0.4x0.4x0.4 [mm3] and projecting the center of each voxel to the individual maps for interpolation. The distance between the projected grid coordinates and the nearest image voxel was used as interpolation weight.
The HCPh dataset will be publicly released at the end of Stage 2 of the corresponding registered report. In addition, the template and individual images will be released. The code used for template generation is openly available at https://github.com/acionca/hcph-template.
Results:
A high-SNR, multivariate template of a single-human brain's anatomy. The template shows sharp boundaries between the different brain tissue and high definition of cortical structures. Furthermore, SNR within the whole brain, the white matter and the gray matter is increased when compared to the original images.
Precise single-subject brain parcellation of the white matter and pial surfaces of the brain.Leveraging the high similarity between individual images resulted in sharper tissue probability coupled with precise surface definition without abnormality within the mesh.
Distance-weighted interpolation of the input images to build a super-resolution template. The approach to super-resolution interpolation, which considers the distance between projected coordinates, provided improved sharpness and contrast between the brain tissues.
Precise single-subject brain parcellation of the white matter and pial surfaces of the brain.Leveraging the high similarity between individual images resulted in sharper tissue probability coupled with precise surface definition without abnormality within the mesh.
Distance-weighted interpolation of the input images to build a super-resolution template. The approach to super-resolution interpolation, which considers the distance between projected coordinates, provided improved sharpness and contrast between the brain tissues.
·Figure 1. Anatomical cut of the improved T1w Colin 27 template (top) along with our T1w (center) and T2w (bottom) templates. All brains were aligned to the same coordinate system.
·Figure 2. Brain intensity histogram (left) for each individual T1w image (blue) and for the T1w template (orange) along with T1w (top-right) and T2w (bottom-right) tissue SNR comparisons.
Conclusions:
We provide a high-definition template derived from a single healthy individual and generated from T1w and T2w contrasts at 3T. This initiative is an effort to grow interest in densely sampled datasets and to improve mapping of the individual human brain.
Modeling and Analysis Methods:
Image Registration and Computational Anatomy
Multivariate Approaches 2
Neuroinformatics and Data Sharing:
Brain Atlases 1
Keywords:
MRI
Open Data
Open-Source Code
Spatial Normalization
STRUCTURAL MRI
White Matter
Other - Template
1|2Indicates the priority used for review
Provide references using author date format
Evans, A. C., A. L. Janke, D. L. Collins, and S. Baillet. 2012. "Brain templates and atlases." NeuroImage 62 (2): 911–922. doi: 10.1016/j.neuroimage.2012.01.024.
Naselaris, T., E. Allen, and K. Kay. 2021. "Extensive sampling for complete models of individual brains." Current Opinion in Behavioral Sciences 40: 45–51. doi: 10.1016/j.cobeha.2020.12.008.
Aubert-Broche, B., A. C. Evans, and L. Collins. 2006. "A new improved version of the realistic digital brain phantom." Neuroimage 32 (1): 138–145. doi: 10.1016/j.neuroimage.2006.03.052.
Provins, C., et al. 2023. "Reliability characterization of MRI measurements for analyses of brain networks on a single human [Registered Report Stage 1 manuscript]." doi: 10.6084/m9.figshare.19579873.v1.
Tustison, N. J., et al. 2010. "N4ITK: Improved N3 Bias Correction." IEEE Transactions on Medical Imaging 29 (6): 1310–1320. doi: 10.1109/TMI.2010.2046908.
Manjón, J. V., P. Coupé, L. Martí-Bonmatí, D. L. Collins, and M. Robles. 2010. "Adaptive non-local means denoising of MR images with spatially varying noise levels." J Magn Reson Imaging 31 (1): 192–203. doi: 10.1002/jmri.22003.
Dale, A. M., B. Fischl, and M. I. Sereno. 1999. "Cortical surface-based analysis. I. Segmentation and surface reconstruction." Neuroimage 9 (2): 179–194. doi: 10.1006/nimg.1998.0395.
Avants, B. B., N. J. Tustison, G. Song, P. A. Cook, A. Klein, and J. C. Gee. 2011. "A reproducible evaluation of ANTs similarity metric performance in brain image registration." Neuroimage 54 (3): 2033–2044. doi: 10.1016/j.neuroimage.2010.09.025.