An atlas of the functional specialization of the human brain's white matter
Presented During:
Monday, June 24, 2024: 5:45 PM - 7:00 PM
COEX
Room:
ASEM Ballroom 202
Poster No:
2208
Submission Type:
Abstract Submission
Authors:
Marcela Ovando-Tellez1, Chris Foulon2, Michel Thiebaut de Schotten3
Institutions:
1Groupe d'Imagerie Neurofonctionnelle (GIN), Institut des Maladies Neurodegeneratives-UMR 5293, CNRS, Bordeaux, France, 2Groupe d'Imagerie Neurofonctionnelle (GIN), Institut des Maladies Neurodegeneratives-UMR 5293, CNRS, Bordeaux, France, 3Groupe d’Imagerie Neurofonctionnelle, Institut des Maladies Neurodégénératives- UMR 5293, CNRS, CEA, Bordeaux, France
First Author:
Marcela Ovando-Tellez
Groupe d'Imagerie Neurofonctionnelle (GIN), Institut des Maladies Neurodegeneratives-UMR 5293, CNRS
Bordeaux, France
Groupe d'Imagerie Neurofonctionnelle (GIN), Institut des Maladies Neurodegeneratives-UMR 5293, CNRS
Bordeaux, France
Co-Author(s):
Chris Foulon
Groupe d'Imagerie Neurofonctionnelle (GIN), Institut des Maladies Neurodegeneratives-UMR 5293, CNRS
Bordeaux, France
Groupe d'Imagerie Neurofonctionnelle (GIN), Institut des Maladies Neurodegeneratives-UMR 5293, CNRS
Bordeaux, France
Michel Thiebaut de Schotten
Groupe d’Imagerie Neurofonctionnelle, Institut des Maladies Neurodégénératives- UMR 5293, CNRS, CEA
Bordeaux, France
Groupe d’Imagerie Neurofonctionnelle, Institut des Maladies Neurodégénératives- UMR 5293, CNRS, CEA
Bordeaux, France
Introduction:
Cognitive functions such as memory, attention and language are critical for our survival and success as a species. They rely on cortical networks connected by bundles of axons (i.e., white matter) which, when disconnected, can lead to different disorders. While tremendous progress in our anatomical description of white matter has been achieved in the last 20 years [1], the relationship between brain connections and the emergence of cognitive functions remains elusive. This proposal aims to unveil the functional specialization of the white matter in the brain in a data-driven way [2].
Methods:
The data for 110 participants were derived from the HCP 7T release. This dataset comprises alternate fMRI sessions of resting and video-watching that have already undergone preprocessing and registration to the MNI152 reference space. The dataset was divided into two groups of 55 subjects each, with the first group designated as the discovery group and the second group as the replication group.
The functionnectome method applied to naturalistic videos : Throughout each session, the variation in the activation for each participant and voxel over time was z-normalized. We estimated the contribution of the associative tracts of the white matter circuits to the cortical variation during the video-watching, by projecting the data onto a recently developed method called functionnectome (i.e., functionnectome time series)[3]. The functionnectome is generated by projecting the fMRI signal from grey matter voxels to the white matter and weighing the signal by the probability of structural connectivity between each grey matter voxel and the rest of the brain. Subsequently, for each group, each time point of the functionnectome time series underwent a one-sample t-test to reveal the significant variation of each voxel in the brain in response to the videos.
White matter parcellation based on the functional organization : The profiles of the time course variation for each brain voxel were entered into a Uniform Manifold Approximation and Projection (UMAP) embedding design, where voxels with similar covariation clustered together, and voxels with different profiles were placed farther apart. Finally, clusters in that space were identified using the HDBscan clustering algorithm. Different parameters for the HDBscan clustering algorithm were evaluated, and we selected the algorithm with the highest Density Based Cluster Validity (DBCV; [4]) and projected it back onto the brain to provide the first division of white matter based on functional activations.
Validation of the results : The UMAP analysis was also computed in the replication group. We computed the Euclidean distance matrix between 1000 random values in the UMAPs of the discovery and replication groups [5]. We calculated the Pearson correlation coefficient between these two distance matrices to validate the UMAP reproducibility.
The functionnectome method applied to naturalistic videos : Throughout each session, the variation in the activation for each participant and voxel over time was z-normalized. We estimated the contribution of the associative tracts of the white matter circuits to the cortical variation during the video-watching, by projecting the data onto a recently developed method called functionnectome (i.e., functionnectome time series)[3]. The functionnectome is generated by projecting the fMRI signal from grey matter voxels to the white matter and weighing the signal by the probability of structural connectivity between each grey matter voxel and the rest of the brain. Subsequently, for each group, each time point of the functionnectome time series underwent a one-sample t-test to reveal the significant variation of each voxel in the brain in response to the videos.
White matter parcellation based on the functional organization : The profiles of the time course variation for each brain voxel were entered into a Uniform Manifold Approximation and Projection (UMAP) embedding design, where voxels with similar covariation clustered together, and voxels with different profiles were placed farther apart. Finally, clusters in that space were identified using the HDBscan clustering algorithm. Different parameters for the HDBscan clustering algorithm were evaluated, and we selected the algorithm with the highest Density Based Cluster Validity (DBCV; [4]) and projected it back onto the brain to provide the first division of white matter based on functional activations.
Validation of the results : The UMAP analysis was also computed in the replication group. We computed the Euclidean distance matrix between 1000 random values in the UMAPs of the discovery and replication groups [5]. We calculated the Pearson correlation coefficient between these two distance matrices to validate the UMAP reproducibility.
Results:
The UMAP analysis was replicated in the replication group, demonstrating consistency in the distribution of data points (see Figure 1). The Pearson correlation coefficient, quantifying the similarity in Euclidean distance between UMAP embeddings in both groups, was 0.98. The highest DBCV for the HDBscan clustering algorithms was 0.36 (see Figure 2A). This analysis yielded a robust parcellation comprising 83 parcels, covering a substantial 64% of the white matter voxels (see Figure 2B). These findings underscore the reproducibility of our UMAP design between groups, providing a reliable foundation for interpreting and generalizing the identified white matter functional organizations.
·UMAP embedding spaces for the discovery and replication groups. In the discovery group, the data were randomly entered into UMAP to illustrate that the observed patterns are not a result of randomness
·The HDBscan clustering algorithms applied to our data provide the first division of white matter based on functional activations.
Conclusions:
By combining fMRI functional signals with white matter circuit anatomy, we linked brain activation to the structure of neural connections. This innovative methodology enabled us to create a comprehensive white matter parcellation tuned to brain activations and, accordingly, functional specialization.
Modeling and Analysis Methods:
Segmentation and Parcellation
Other Methods
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity 2
Neuroinformatics and Data Sharing:
Brain Atlases 1
Keywords:
White Matter
Other - brain parcellation
1|2Indicates the priority used for review
Provide references using author date format
[2]Finn, E. S., & Bandettini, P. A. (2021). Movie-watching outperforms rest for functional connectivity-based prediction of behavior. NeuroImage.
[3]Nozais, V., Forkel, S. J., Foulon, C., Petit, L., & Thiebaut de Schotten, M. (2021). Functionnectome as a framework to analyse the contribution of brain circuits to fMRI. Communications Biology.
[4]Moulavi, D., Jaskowiak, P.A., Campello, R.J., Zimek, A. and Sander, J., 2014. Density-Based Clustering Validation. In SDM (pp. 839-847).
[5]Pacella, V., Nozais, V., Talozzi, L., Forkel, S. J., & Thiebaut de Schotten, M. (submitted). Unravelling the fabric of the human mind: the brain-cognition space.