Connectome Embedding in Multidimensional Graph Spaces
Author Information
Author(s): Mach Mathieu, Amico Enrico, Liégeois Raphaël, Preti Maria Giulia, Griffa Alessandra, Van De Ville Dimitri, Pedersen Mangor
Primary Institution: Neuro-X Institute, Ecole Polytechnique Fédérale De Lausanne (EPFL), Geneva, Switzerland
Hypothesis
Can brain networks be effectively analyzed using high-dimensional graph spaces defined by multiple nodal properties?
Conclusion
The study proposes a new framework for quantifying network features in high-dimensional spaces, revealing new properties of brain networks.
Supporting Evidence
- Machine learning analysis achieved 80-86% accuracy in classifying brain regions.
- Nodal properties were found to be correlated at both whole-brain and subnetwork levels.
- The largest gain in machine learning accuracy was observed when moving from 2D to 3D space.
- Euclidean distances between nodes were highest in the default mode network and frontoparietal areas.
Takeaway
This study looks at how brain connections can be better understood by using a new way to measure them in many dimensions, helping us see how different parts of the brain work together.
Methodology
The study used graph theory to analyze brain networks in high-dimensional spaces defined by up to 10 nodal properties, employing machine learning for classification.
Potential Biases
Potential biases may arise from the selection of nodal properties and the use of specific machine learning models.
Limitations
The study's accuracy could be improved with K-fold cross-validation and a larger sample size.
Participant Demographics
100 healthy unrelated subjects from the Human Connectome Project.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website