Graphs and the idiographic brain

Abstract

This scientific commentary refers to ‘Integrating direct electrical brain stimulation with the human connectome’ by Coletta et al. (https://doi.org/10.1093/brain/awad402).

Figure 1

Heterogeneity in structural connectivity. A total of 178 tractography datasets from healthy controls were acquired with 7 T MRI and encoded as graphs where nodes represent grey matter voxels and edges a streamline connection between them. A connectivity image was generated for each subject using the number of edges per node as a voxel-wise measure of the degree of structural connectivity. Each image was re-represented as a 100-dimensional vector using Gaussian random projection and then embedded in a two-dimensional latent space with t-distributed stochastic neighbour embedding (see central plot). Agglomerative clustering was used to derive four clusters (seen as distinct colours in the embedding). The mean connectivity image for each cluster is shown projected onto the surface of a template brain. Even in this simplified representation, each cluster shows variation that is particularly marked in the frontal lobe and cerebellum. Causal structure-function maps that can leverage connectomes more closely related to each individual—here from members in the nearest cluster—are likely to provide better downstream fidelity and reduced bias as compared to those that use a normative connectome.