Defining functional brain networks using unsupervised density-peak clustering
| Autor: | Ossmy O, Mukamel R |
|---|---|
| Rok vydání: | 2020 |
| Předmět: | |
| DOI: | 10.31234/osf.io/hkrzw |
| Popis: | Background: Parcellating the human brain into areas based on their neural connectivity is essential for understanding the functional organization of neural networks. New Method: We adapted density-peak clustering to identify functional neural networks in individuals without the need to select seed regions. We first assess the similarity between each pair of voxels based on their activation time-courses and then aggregate the voxels based on the assumption that cluster centers are dense (have high similarity with many voxels) and are at large distance from other high-density voxels. This data-driven approach allows intuitive selection of cluster centroids in individual subjects.Results: We applied our approach on resting-state data of individual subjects. Although similar networks across subjects were identified, there was large variability in the number of networks and their anatomical distribution between subjects. Manipulating the main free parameter of the model (density level threshold) revealed a hierarchic representation in which large clusters are divided to smaller sub-clusters when decreasing the threshold.Comparison with Existing Method: To date, most connectiviy-based parcellations begin with selecting an initial seed region and are therefore limited and heavily reliant on prior theoretical knowledge. Existing methods also require many pre-defined parameters and were usually used at the group level. Conclusions: Adapting density-peak clustering algorithm to neural data has potential implications for understanding individual differences in functional networks without pre-determining the number of networks or functional/anatomical definition of a seed region. This data-driven approach may pave the way to deeper investigation of the brain structure-function relationship within individual humans. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|