The brain has undeniably been the most enigmatic organ of the human body. To further our understanding of this mysterious organ, a talk titled “Controllability of Functional Brain Networks” was organized on 2nd September 2021 as a part of the third AI impact seminar. The talk was delivered by Professor Ramkrishna Pasumarthy who is an Associate Professor at IIT Madras.

Prof. Pasumarthy started the talk by saying that for a long time most brain modelling was questionnaire-based and only now the availability of huge amounts of FMRI data is driving advances in this field. Talking about his research interest, he said that his lab focuses on control theory and network science.

Next, he introduced various components of complex networks and explained nodes and edges which are subparts of the system. Further, he went on to explain a bit about network interpretation of the human brain. He said that the brain is an organ and it is difficult to visually interpret it as a connection of little pieces. The brain has peaks and troughs which help define parts that have different kinds and varying densities of neurons and depending on the type and density of neurons these parts can have individual functions thus each of these regions can be considered as nodes. He further told that standard diffusing imaging techniques show some behaviour of neuronal axons which define how information transfer happens between these regions and from a graph point of view one can consider them as edges. He then enumerated brain regions along with their functions.

Next, he listed questions to ponder over in this area such as how various regions interact with each other? Are they symmetric? Are interactions fixed or vary with time? Are these interactions consistent among several subjects? Do interactions and hence cognitive abilities change based on cognitive load/tasks? Does the structure remain the same while sleeping, watching sport etc?

He then explained network controllability and said that it is of interest to assign a value to each node, or set of nodes, which quantifies the influence of each node or set of nodes over the entire network. He explained that it was expected that nodes with high average controllability are the ones which are expected to have a larger (control-) influence over the network. He then listed the questions on this problem like what are the state space models for estimating causal interaction among various regions of interest?, how to identify weighted directed graphs?, To know if there exist a causal control hubs?, and to identify the nodes which exhibit higher controllability and is controllability task dependent. Next, he described that controllability measures the ability to perturb a system from given initial state to random target states and commented that hubs are highly connected regions important for integration of activity across brain regions.

Describing his research work, he said that they took resting state data from the Human Connectome project with time series data with 26 nodes and 1180 sample points of 100 healthy subjects which were clustered to 121 nodes based on regions and location names. From this data, the team generated 10 snapshots for the temporal network and used the methods of thresholding to generate the system adjacency matrix. The team wanted to find out the set of nodes that can control the entire network and the research showed that the default mode network is the one which is most influential. Each of these clusters can control all 96 subjects. In the resting state network, the default mode network was found to be the most active and influential one. He reasoned this saying that we are not doing any task in resting state.

He next talked about his research related to working memory tasks and told that cognitive control processes associated with working memory play an essential role in development and impairments in these processes are a key component of cognitive dysfunction in many psychiatric disorders like schizophrenia and attention deficit hyperactivity disorder. He said that the knowledge about the dynamic casual and asymmetric interactions between distributed brain regions during working memory task performance is present due to small sample sizes which are unreliable and so his research team decided to work on this problem. Prof. Ramkrishna’s team selected 737 subjects from Human connectome data and from time series they identified multivariate state space models and looked at each subjects for three things: how to quantify causal interactions between these regions?, Can they actually classify between tasks? and what happens to controllability when they do a certain task? They restricted themselves to 0 back data and found that salient network and Fronto-parietal network regions of interest showed significantly greater activation in the 2- back compared to 0-back. Also, they found that the default mode network showed significantly reduced activity in the 2- back condition. Their results reinforced that the AI, a key node in the Salient network, is engaged in a wide range of cognitive control and it has an important role in detecting salient external signals, relocating cognitive resources and directing goal-directed behaviour. The team also found that network controllability was higher in 0-back compared to the 2-back tasks, network controllability decreases with working memory load and the salient network has the highest controllability and not the default mode network.

He concluded his talk by saying that controllability does not depend on how nodes are interconnected to each other but is task-dependent. Discussing his future plans, he said he wishes to characterize the entire brain network and understand functionalities and discover neurological markers using a deep learning framework and large-scale brain circuit dynamics features in order to distinguish patients with autism from healthy individuals, predict symptom severity for patients with autism and predict progression of symptoms in patients with autism. The talk piqued the interest of listeners and was followed by a question-answer round.

The video is available on our YouTube channel: Link.

Keywords

Brain Modelling, Cognitive Control