Back to Projects

Image Computing and Computational Modelling (pre-2019)

Biologically interpretable models for brain disorders

Project ID: 2015_311

Student: Jessica Dafflon

1st supervisor: Federico Turkheimer, King’s College London
2nd supervisor: Peter Hellyer, Imperial College London

Recently, there has been increased interest in understanding the brain as a “critical” system poised between order and chaos. Theoretically, such scale-free ‘critical’ dynamics optimize properties that may be behaviourally useful. However, it is unclear how these dynamical measures relate to neurobiology in the human brain.

Advances in neuroimaging techniques such as MR-Diffusion-based tractography and fMRI, have afforded the possibility of detailed examination of ‘human connectome’ in terms of both function (dynamic) and structure (stable over short timescales). The emergence of multiple ‘functional connectivity’ states from one structural topology is an example of ‘functional multiplicity’ and suggests that not only is the structure of the brain important in predicting functional outcome, but also the local dynamics of brain regions.

Computational models to simulate brain dynamics range from simple oscillator models to simulations of pools of excitatory and inhibitory neurons (such as the Wilson-Cowan model and the more modern Dynamical Mean Field (DMF) model). A common approach is to consider alerted structural topology, in terms of overall connectivity strength and graph theoretic motifs. Typically, the dynamics of the system can then be compared to the generative accuracy of the model in terms of modelling ‘resting state’ behaviour. While much is known about the changes to network topology in the emergence of critical regimes, relatively little is known about the neuropharmacological/pathological factors that influence these regimes.

We plan to adapt this framework to examine the influence of distributed neurotransmitter systems on network dynamics. In more complex models such as the Wilson-Cowan and DMF models we have the ability to selectively modulate the activity of individual sub groups of neuronal populations at each individual node within the network. This enables us to realistically change the dynamics of local regions of the model in response to pharmacological intervention or modulate the activity of local pools of neurons e.g. neuropsychiatric / developmental disorders.

In this project, we will evaluate a range of computational models, grounded in physiological constraints and how they predict empirical functional connectivity and empirical measures of critical brain dynamics in the healthy brain. We will then adapt these models to examine changes to functional connectivity in neuropsychiatric disorder or pharmacological challenge using two complementary approaches:
a) We will adapt the local dynamics of these models (neurotransmitter balance, activity excitatory and inhibitory pools) according to well-tested hypotheses of disease / neurotransmitter activity within the brain – driven by localization of neurotransmitter systems using PET / mRNA mapping developed from the Allen Brain Atlas.
b) We will employ data driven approaches using multivariate machine learning and optimization approaches, to lesion normative models, and compare to functional connectivity from pharmacological / psychiatric datasets.

These generative models, likely will produce a rich source of data, which can further be combined with empirical observations of complex population dynamics to predict outcome (prognosis and diagnosis). The final aim of the project would be to combine the insights generated throughout the work to develop interpretable machine learning driven models of outcome in both pharmacological intervention and disease.

[1] Hellyer PJ, Shanahan M, Scott G, et al., 2014, The Control of Global Brain Dynamics: Opposing Actions of Frontoparietal Control and Default Mode Networks on Attention, Journal of Neuroscience, Vol:34, ISSN:0270-6474, Pages:451-461
[2] Expert P, Lambiotte R, Chialvo DR, Christensen K, Jensen HJ, Sharp DJ, Turkheimer F. Self-similar correlation function in brain resting-state functional magnetic resonance imaging. J R Soc Interface. 2011 Apr 6;8(57):472-9

Back to Projects