Inter-Individual Variability in Brain Function Captured by a Siamese Neural Network

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
Neuroscience has focused on group differences for a long time. However, individual variability can be important for explaining and predicting brain disorders and age-related changes. Here, we will investigate these inter-individual differences by using data from the Human Connectome Project. Two resting-state fMRI time series will be inputted to a Siamese neural network that will learn whether the two inputs are from the same person, or from two differing persons. We show that it can do this with an accuracy of 93%. The features learned by the network showed clear links to known large-scale brain networks. We found 1) a higher-level network component with correlations to lower-level networks, 2) an intra-network component, and 3) a higher-level network component without correlations to lower-level networks. The first component, in particular the default mode network, the dorsal attention network and the cinguloopercular network, showed an association with fluid intelligence, speed of processing, language comprehension and emotion recognition. The intra-network component showed a link with cognitive tests such as episodic memory, cognitive flexibility and speed, language, intelligence and emotion recognition. So, we showed that we can identify a low-dimensional representation of inter-individual variability of brain function by using a Siamese neural network, where the features can be linked to large-scale neural network and cognitive function. Keywords: functional connectivity, Siamese neural networks, resting state, fMRI
Faculteit der Sociale Wetenschappen