Farrugia, N. e. | Search bioRxiv

bioRxiv2021

Deep learning models of cognitive processes constrained by human brain connectomes

Decoding cognitive processes from recordings of brain activity has been an active topic in neuroscience research for decades. Traditional decoding studies focused on pattern classification in specific regions of interest and averaging brain activity over many trials. Recently, brain decoding with graph neural networks has been shown to scale at fine temporal resolution and on the full brain, achieving state-of-the-art performance on the human connectome project benchmark. The reason behind this success is likely the strong inductive connectome prior that enables the integration of distributed patterns of brain activity. Yet, the nature of such inductive bias is still poorly understood. In this work, we investigate the impact of the inclusion of multiple path lengths (through high-order graph convolution), the homogeneity of brain parcels (graph nodes), and the type of interactions (graph edges). We evaluate the decoding models on a large population of 1200 participants, under 21 different experimental conditions, acquired from the Human Connectome Project database. Our findings reveal that the optimal choice for large-scale cognitive decoding is to propagate neural dynamics within empirical functional connectomes and integrate brain dynamics using high-order graph convolutions. In this setting, the model exhibits high decoding accuracy and robustness against adversarial attacks on the graph architecture, including randomization in functional connectomes and lesions in targeted brain regions and networks. The trained model relies on biologically meaningful features for the prediction of cognitive states and generates task-specific graph representations resembling task-evoked activation maps. These results demonstrate that a full-brain integrative model is critical for the large-scale brain decoding. Our study establishes principles of how to effectively leverage human connectome constraints in deep graph neural networks, providing new avenues to study the neural substrates of human cognition at scale.

neuroscience↗

bioRxiv2021

Representational learning of brain responses in executive function and higher-order cognition using deep graph convolutions

Brain decoding aims to infer human cognition from recordings of neural activity using modern neuroimaging techniques. Studies so far often concentrated on a limited number of cognitive states and aimed to classifying patterns of brain activity within a local area. This procedure demonstrated a great success on classifying motor and sensory processes but showed limited power over higher cognitive functions. In this work, we investigate a high-order graph convolution model, named ChebNet, to model the segregation and integration organizational principles in neural dynamics, and to decode brain activity across a large number of cognitive domains. By leveraging our prior knowledge on brain organization using a graph-based model, ChebNet graph convolution learns a new representation from task-evoked neural activity, which demonstrates a highly predictive signature of cognitive states and task performance. Our results reveal that between-network integration significantly boosts the decoding of high-order cognition such as visual working memory tasks, while the segregation of localized brain activity is sufficient to classify motor and sensory processes. Using twin and family data from the Human Connectome Project (n = 1,070), we provide evidence that individual variability in the graph representations of working-memory tasks are under genetic control and strongly associated with participants in-scanner behaviors. These findings uncover the essential role of functional integration in brain decoding, especially when decoding high-order cognition other than sensory and motor functions. TeaserO_LIModelling functional integration through graph convolution is a necessary step towards decoding high-order human cognition. C_LI Significance statementOver the past two decades, many studies have applied multivariate pattern analysis to decode what task a human participant is performing, based on a scan of her brain. The vast majority of these studies have however concentrated on select regions and a specific domain, because of the computational complexity of handling full brain data in a multivariate model. With the fast progress in the field of deep learning, it is now possible to decode a variety of cognitive domains simultaneously using a full-brain model. By leveraging our prior knowledge on brain organization using a graph-based model, we uncovered different organizational principles in brain decoding for motor execution and high-order cognition by modelling functional integration through graph convolution.

neuroscience↗