Poster No:
1726
Submission Type:
Abstract Submission
Authors:
Byeongwook Lee1, Louis Rouillard2, Luca Ambrogioni3, Srikanth Ryali1, Demian Wassermann2, Vinod Menon1
Institutions:
1Stanford University, Palo Alto, United States, 2Université Paris-Saclay, Inria, CEA, Paris, France, 3Donder's Institute of Cognition, Radboud University, Nijmegen, Netherlands
First Author:
Co-Author(s):
Luca Ambrogioni
Donder's Institute of Cognition, Radboud University
Nijmegen, Netherlands
Vinod Menon
Stanford University
Palo Alto, United States
Introduction:
Working memory, a fundamental component of human cognition, encompasses the ability to retain and manipulate information in the mind for various cognitive tasks, a function central to all reasoning, comprehension, and learning [1, 2]. While previous research has made significant strides in identifying key brain regions and probing interactions within limited sets of brain regions, a comprehensive understanding of brain-wide dynamic orchestration of distributed brain regions during working memory has remained elusive. This gap in knowledge is particularly notable in the context of analyzing causal control circuits at the whole-brain level, which are crucial for understanding how specific brain nodes or networks assert control in human working memory through asymmetric influences.
Methods:
Here we bridge this gap by developing a novel Multivariate Dynamic Systems Identification [3]-hybrid Variational Bayes (MDSI-hVB) technique. This novel computational framework is tailored to capture whole-brain asymmetric directed interactions, addressing pivotal challenges in dynamic causal modeling within cognitive neuroscience. These challenges encompass the development of computational methods for large-scale network analysis, modeling regional variations in the hemodynamic response function (HRF), and deciphering context-dependent brain-wide causal interactions. We analyzed brain-wide dynamic causal interactions in a large (N=737) sample of participants from the Human Connectome Project [4] who performed a n-back working memory task during fMRI scanning. DiFuMo atlas [5] was used to extract task fMRI time series for subsequent analyses as it offers fine-grained brain-wide functional modes for both cortical and subcortical areas and features multi-dimensional functional networks. We utilized time series data obtained from 128 dimensions for the main analyses and employed time series data extracted from both 64 and 256 dimensions to assess the scalability and robustness of MDSI-h-VB in estimating whole-brain causal interactions.
Results:
Utilizing MDSI-hVB, we explored intricate brain-wide dynamic causal and asymmetric interactions during working memory. A key aspect of our approach was capturing the heterogeneity of the HRF across different brain regions. This capability allowed our MDSI-hVB model to more accurately represent latent neural activity, reflecting underlying neural dynamics rather than mere blood flow changes. Our MDSI-h-VB model identified asymmetric connectivity patterns exhibiting significant dynamic causal interactions under both the 0-back and 2-back task conditions (p<0.01, FDR-corrected, two-side paired t-test). Notably, load-dependent causal interactions distinguished the task conditions (2-back vs. 0-back) with remarkable accuracy (classification accuracy=98%, p<0.01, permutation test). This high level of accuracy underscores the effectiveness of our approach in capturing brain-wide load-dependent causal circuit dynamics during working memory. Additionally, our MDSI-h-VB model reliably predicted brain-behavior relationships. A critical finding of our study was the identification of directed causal outflows during the working memory task. We discovered that the anterior insula (AI) and the middle frontal gyrus (MFG) function as pivotal outflow and inflow nodes, respectively, at the whole-brain level. This discovery is significant, as it highlights the specific roles individual brain regions play in orchestrating the complex dynamics of working memory processes. Lastly, our methodological framework demonstrated remarkable scalability, successfully extending the analysis to encompass multiple dimensions encompassing 64, 128, and 256 regions.
Conclusions:
Our study not only unravels brain-wide dynamic circuit mechanisms underpinning human working memory but also opens new avenues for exploring context-dependent dynamic causal networks at the whole-brain level, offering deeper insights into dynamic networks that drive human cognition.
Higher Cognitive Functions:
Executive Function, Cognitive Control and Decision Making
Learning and Memory:
Working Memory
Modeling and Analysis Methods:
Bayesian Modeling
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Keywords:
ADULTS
Computational Neuroscience
Data analysis
FUNCTIONAL MRI
Machine Learning
Modeling
1|2Indicates the priority used for review
Provide references using author date format
1. Menon, Vinod, and Mark D’Esposito. "The role of PFC networks in cognitive control and executive function." Neuropsychopharmacology 47.1 (2022): 90-103.
2. D'Esposito, Mark, and Bradley R. Postle. "The cognitive neuroscience of working memory." Annual review of psychology 66 (2015): 115-142.
3. Ryali, Srikanth, et al. "Multivariate dynamical systems models for estimating causal interactions in fMRI." Neuroimage 54.2 (2011): 807-823.
4. Van Essen, David C., et al. "The WU-Minn human connectome project: an overview." Neuroimage 80 (2013): 62-79.
5. Dadi, Kamalaker, et al. "Fine-grain atlases of functional modes for fMRI analysis." NeuroImage 221 (2020): 117126.