Multiscale subdivision of an anatomical cortical parcellation based on geodesic distance
Poster No:
2015
Submission Type:
Abstract Submission
Authors:
Yarelis Prieto1, Joaquín Molina2, Mónica Otero3, Cecilia Hernández4, Cyril Poupon5, Jean-François Mangin6, WAEL EL-DEREDY7, Pamela Guevara8
Institutions:
1Universidad de Concepción, Chile, Concepción, Región del Bío-Bío, 2Universidad de Concepción, Concepción, Región del Bío-Bío, 3Universidad San Sebastián, Santiago de Chile, Metropolitana, 4Universidad de Concepción, Concepción, Bío-Bío, 5Neurospin, Paris, France, 6Université Paris-Saclay, CEA, CNRS, Neurospin, Gif-sur-Yvette, France, 7UNIVERSIDAD DE VALPARAISO, Valparaíso, Valparaíso, 8Universidad de Concepción, Concepcion, Región del Bio-Bio
First Author:
Co-Author(s):
Introduction:
The brain can be analyzed as a complex network of interacting regions, often referred to as the human connectome [Sporns et al., 2004]. To obtain and study the human connectome, a crucial step is to define the gray matter parcellation, over which the connectivity profiles are calculated. Some studies have focused on obtaining different levels or scales of parcellation, ranging from fine- to coarse-grained parcellations, based on a base parcellation [Moreno-Dominguez et al., 2014], [Molina et al., 2023], [Diez et al., 2015] . However, these prior works face challenges such as considerably variations in size and spatial discontinuity of regions. Additionally, parcellations in these studies are not defined with a multiscale structure. Therefore, we propose a framework for the generation of multiscale cortical parcellations on a surface representation of the cortex based on a group of 79 healthy subjects from the ARCHI database [Schmitt et al., 2012]. The method is based on geodesic distance and hierarchical clustering, and uses a tree partitioning method, depending on the maximum distance between regions in a cluster, to obtain different parcellation scales. Moreover, it efficiently calculates the subject's connectome for each level.
Methods:
Fig. 1 illustrates the proposed method. First, based on [López-López et al., 2020], centroid of each region is calculated for each hemisphere of each subject using the geodesic distance between vertices of the region in the base parcellation. Next, a distance matrix is calculated based on the geodesic distance between the centroids of each pair of regions (Fig. 1A). Then, a mean distance matrix is computed to ensure the results reproducibility and its values are entered into an affinity graph (Fig. 1B) [Molina et al., 2023]. Subsequently, an average-link agglomerative hierarchical clustering is applied on the affinity graph to obtain a dendrogram. A tree partitioning method, depending on the maximum distance (Dm) between regions in a cluster, was used to obtain different parcellation scales [Román et al., 2022] (Fig. 1C). The method establishes and saves the correspondence between all levels of parcellation. Finally, based on the Möller-Trumbore ray-triangle intersection [Möller et al., 2005], for each subject, it efficiently calculates the connectome for each level, using the whole brain tractography dataset and the subject's labeled mesh (Fig. 1D).
·Fig. 1. Proposed method. (A) Centroid for each region and the distance matrices for each subject. (B) Mean matrix, affinity graph and dendrogram. (C) Multiscale parcellation. (D) Connectivity matrices
Results:
We applied the method to Desikan-Killiany atlas [Desikan et al.; 2006] as the base parcellation and obtained four parcellation scales for both symmetrical and non-symmetrical parcellations. Fig. 2A displays the multiscale symmetrical parcellation obtained using a single average matrix for both hemispheres and Dm of 10, 42, 84, and 180 mm from level L4 to L1, respectively. Structural connectivity matrices reveal the expected shape, with symmetrical connections between hemispheres and with more connections within each hemisphere. (Fig. 2B). As the parcels are merged, area of the regions and standard deviation are increasing. The highest STD corresponds, as expected, to level L1, which has few large regions (Fig. 2C).
·Fig. 2. Results. (A) Four levels of the multiscale symmetric parcellation. (B) Structural connectivity matrix for each level. (C) Cortical area for the regions in each parcellation level
Conclusions:
We propose an efficient framework to create a multiscale cortical parcellation with configurable parameters and based on any base parcellation. We developed a generic algorithm which does not assume a symmetrical base parcellation, however, since our base parcellation is symmetrical, we obtained a symmetrical multiscale parcellation by using a single mean distance matrix for both hemispheres. The generated parcels exhibit spatial contiguity and homogeneous sizes. With the adaptive partitioning of the hierarchical tree, we can generate as many levels of parcellation as desired, depending on the number of parcels in the base parcellation and the selection of different values of maximum distance between the regions of a cluster for each level. This framework will be made available and can be applied to different fine-grained parcellations.
Modeling and Analysis Methods:
Segmentation and Parcellation 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity 2
Keywords:
Cortex
STRUCTURAL MRI
Tractography
1|2Indicates the priority used for review
Provide references using author date format
Diez, I. (2015), ‘A novel brain partition highlights the modular skeleton shared by structure and function’, Scientific Reports vol. 5, no. 1.
López-López, N. (2020), 'GEOSP: A parallel method for a cortical surface parcellation based on geodesic distance', in ‘2020 42nd Annual International Conference of the IEEE Engineering in Medicine Biology Society (EMBC)’, pp. 1696–1700.
Molina, J. (2023), 'Group-wise cortical parcellation based on structural connectivity and hierarchical clustering', in ‘18th International Symposium on Medical Information Processing and Analysis’, Vol. 12567, pp. 172–181.
Möller, T. (2005), 'Fast, minimum storage ray/triangle intersection', in ‘ACM SIGGRAPH 2005 Courses’, Association for Computing Machinery, p. 7.
Moreno-Domínguez, D. (2014), ‘A hierarchical method for whole-brain connectivity-based parcellation’, Human Brain Mapping vol. 35.
Román, C. (2022), ‘Superficial white matter bundle atlas based on hierarchical fiber clustering over probabilistic tractography data’, NeuroImage, vol. 262, pp. 119550.
Schmitt, B. (2012), 'Connect/archi: an open database to infer atlases of the human brain connectivity’, vol. 272.
Sporns, O. (2004), ‘Organization, development and function of complex brain networks’, Trends in Cognitive Sciences vol. 8, no.9, pp. 418–425.
Acknowledgements: ANID, Chile: Beca Doctorado Nacional 2022-21210468, FONDECYT 1221665, ANILLO ACT210053, Basal FB0008, and Basal FB210017.