Automatic, whole-brain segmentation of cortical layers for the high-resolution anatomical data at 7T
Poster No:
1842
Submission Type:
Abstract Submission
Authors:
Roman Belenya1, Gabriel Castrillón2, Valentin Riedl3
Institutions:
1Klinikum rechts der Isar, Technical University of Munich, Munich, Bavaria, 2Friedrich-Alexander University, Erlangen, Germany, 3Technical University of Munich, Erlangen, Germany
First Author:
Co-Author(s):
Introduction:
Developments of high-resolution MR imaging at 7 Tesla have allowed researchers to study the brain in unprecedented detail. The increased magnetic field improves the signal-to-noise ratio and the resolution of the images. However, 7T imaging comes with a set of challenges. Among them are magnetic field inhomogeneities and imaging artefacts. They present a common challenge for data preprocessing. Over the years, robust and reproducible preprocessing methods have been well-optimized for standard 3T imaging. However, there are currently no standard optimized pipelines for high-resolution brain images. In this work, we develop a preprocessing pipeline for the 7T hires anatomical data. We focus on performing as few processing steps as possible while maximizing the performance of automatic whole-brain segmentation. Specifically, our end-goal is to maximize the accuracy and reliability of cortical layer segmentation for the analyses of layer fMRI data.
Methods:
We processed the anatomical data of 14 participants from the 7T Siemens MAGNETOM Terra.X machine. We collected the 3D T1-weighted MP2RAGE as well as T2-weighted TSE images. The preprocessing pipeline is similar for both modalities:
Pre-freesurfer
1. Image denoising using the DenoiseImage program in ANTs.
1a. Gibbs ringing artefacts removal with the mrdegibbs program implemented in mrtrix. We run this step only for the T2w images because they are affected by the artefact the most.
2. Bias field correction using SPM tissue-type segmentation.
3. Brain extraction with HD-BET.
Freesurfer
4. Brain surface reconstruction with freesurfer recon-all script. We run the standard processing using the high-resolution (-cm) option and an additional T2w contrast to improve the tissue type segmentation. We entered the combined brain mask from Step 3 using the expert file.
Post-Freesurfer
5. To convert brain surfaces generated by the freesurfer to a voxel space rim file required by LayNii, we use the surf-laynii script by Sriranga Kashyap. The script first expands the grey matter and shrinks the white matter surfaces using the mri_expand tool. The mri_fill tool fills the original, expanded and shrunk surfaces, resulting in a set of masks. Subtracting the smaller from a larger mask estimates the boundary between the white matter and the grey matter and the boundary between the CSF and the white matter in voxel space.
6. We use these boundaries as the input to the LayNii LN2_LAYERS program, which estimates cortical layers using an equivolumetric approach.
Pre-freesurfer
1. Image denoising using the DenoiseImage program in ANTs.
1a. Gibbs ringing artefacts removal with the mrdegibbs program implemented in mrtrix. We run this step only for the T2w images because they are affected by the artefact the most.
2. Bias field correction using SPM tissue-type segmentation.
3. Brain extraction with HD-BET.
Freesurfer
4. Brain surface reconstruction with freesurfer recon-all script. We run the standard processing using the high-resolution (-cm) option and an additional T2w contrast to improve the tissue type segmentation. We entered the combined brain mask from Step 3 using the expert file.
Post-Freesurfer
5. To convert brain surfaces generated by the freesurfer to a voxel space rim file required by LayNii, we use the surf-laynii script by Sriranga Kashyap. The script first expands the grey matter and shrinks the white matter surfaces using the mri_expand tool. The mri_fill tool fills the original, expanded and shrunk surfaces, resulting in a set of masks. Subtracting the smaller from a larger mask estimates the boundary between the white matter and the grey matter and the boundary between the CSF and the white matter in voxel space.
6. We use these boundaries as the input to the LayNii LN2_LAYERS program, which estimates cortical layers using an equivolumetric approach.
Results:
We run the pipeline on data from 14 subjects. On average, the pre-freesurfer pipeline doubles the spatial SNR and CNR within the grey and white matter masks for T1w and T2w images. Combining the preprocessed contrasts allows for more accurate placement of brain cortical surfaces and estimation of cortical layers. We demonstrate an example of processing steps and their effect on the images in Figure 1.
·Example of preprocessing steps. Orig) Original T1w MP2RAGE image. 1) Result of the denoising step. 2) Result of the bias correction step. 3) Result of the brain extraction. 4) Tissue type segmentation
Conclusions:
The accuracy of brain segmentation is crucial in analyzing neuroimaging data. While manual segmentations can be most accurate, they are time-consuming, especially on the whole-brain level. The existing automatic segmentation tools are usually optimized for images from 3T scanners and cannot be used out of the box for high-resolution 7T data. Thus, it is essential to configure the preprocessing pipelines to maximize the accuracy of automatic whole-brain segmentation. Here, we describe the processing pipeline that optimized the performance of cortical layer segmentation for our anatomical data. Accurate automatic layer segmentation can dramatically increase the accessibility of high-resolution data analyses. This work will assist researchers in developing analysis methods for the high-resolution neuroimaging data.
Modeling and Analysis Methods:
Image Registration and Computational Anatomy 1
Methods Development
Segmentation and Parcellation
Neuroinformatics and Data Sharing:
Workflows 2
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Acquisition
Cortical Layers
HIGH FIELD MR
MRI
Segmentation
STRUCTURAL MRI
1|2Indicates the priority used for review
Provide references using author date format
Isensee F. (2019). Automated brain extraction of multi-sequence MRI using artificial neural networks. Human Brain Mapping, 1–13
Sriranga Kashyap. Surf-laynii. https://github.com/srikash/surf_laynii
Tournier, J.-D. (2019). MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. NeuroImage, 202, 116137
Tustison, N. J. (2014). Large-scale evaluation of ANTs and FreeSurfer cortical thickness measurements. NeuroImage, 99, 166–179