Modelling variability in functional brain networks using deep generative models
Poster No:
1180
Submission Type:
Abstract Submission
Authors:
Rukuang Huang1, Mark Woolrich2, Chetan Gohil1
Institutions:
1OHBA, Department of Psychiatry, University of Oxford, Oxford, N/A, 2University of Oxford, Oxford, Oxon
First Author:
Co-Author(s):
Introduction:
There is a growing interest in studying the dynamics of functional networks, which have previously been linked to cognition, demographics and disease states. The sliding window approach is one of the most common approaches to compute dynamic functional networks. However, it cannot reliably detect cognitively relevant and transient temporal changes at the time scales of fast cognition, i.e. on the order of 100 milliseconds. In combination with electrophysiological data, these fast temporal changes of functional connectivity (FC) can be identified with generative modelling based methods such as the HMM (Hidden Markov Models) and DyNeMo (Dynamic Network Modes). We attempted to address two of the limitations of these generative models.
Methods:
Here we introduce two key limitations of current generative models and describe how we solve them. Firstly, time-varying estimates of power and FC are calculated under the assumption that they share the same dynamics but there is no principled basis for this assumption. We propose Multi-Dynamic Network Modes (M-DyNeMo) that allows for the possibility that power and the FC are uncoupled.
Secondly, most of the current methods for estimating (both static and dynamic) FC, including the sliding window approach and generative models, assume the same set of functional networks for all sessions, i.e. the networks are estimated at the group level. This does not allow for the discovery of, nor benefit from, subpopulation structure in the data. We propose the use of embedding vectors (c.f. word embedding in Natural Language Processing) to explicitly model individual sessions.
Secondly, most of the current methods for estimating (both static and dynamic) FC, including the sliding window approach and generative models, assume the same set of functional networks for all sessions, i.e. the networks are estimated at the group level. This does not allow for the discovery of, nor benefit from, subpopulation structure in the data. We propose the use of embedding vectors (c.f. word embedding in Natural Language Processing) to explicitly model individual sessions.
Results:
In Figure 1, we see that M-DyNeMo can extract uncoupled network dynamics for power and FC from amplitude envelope data. With an evoked response analysis by epoching the network dynamics, we show that the two dynamics respond differently to a visual task, which can be shown further to induce a significant difference in the coupling between the dynamics. These findings can be reproduced with data from different subjects, different datasets and different parcellations. We also used simulation to validate that if the underlying dynamics of power and FC are indeed coupled, M-DyNeMo will infer coupled dynamics, meaning the uncoupled dynamics we see in real data is not an artefact of the model but genuine stucture in the data. Finally, a graphical representation of generative model of M-DyNeMo is illustrated on the right.
In Figure 2, we show results of applying DIVE - an extension of DyNeMo with the embedding technique, on two real datasets. In the first dataset, each of the 19 subjects are scanned 6 times. In the top right panel, each recording session is assigned an embedding vector and the session-pairwise cosine distance of the embeddings are plotted. A clear block diagonal structure can be seen, meaning sessions of the same subjects are grouped together in the embedding space in an unsupervised manner. In the bottom panel, subject recordings from two different datasets are used. If we post-hoc colour the inferred embedding vectors with age-range and original dataset of the subjects, we see that age and dataset information are encoded simultaneously in the embedding space in different directions. Furthermore, we can pick vectors corresponding to an old and a young subject and generate their respective connectivity networks and study the differences in spatial configuration as well as frequency contents.
In Figure 2, we show results of applying DIVE - an extension of DyNeMo with the embedding technique, on two real datasets. In the first dataset, each of the 19 subjects are scanned 6 times. In the top right panel, each recording session is assigned an embedding vector and the session-pairwise cosine distance of the embeddings are plotted. A clear block diagonal structure can be seen, meaning sessions of the same subjects are grouped together in the embedding space in an unsupervised manner. In the bottom panel, subject recordings from two different datasets are used. If we post-hoc colour the inferred embedding vectors with age-range and original dataset of the subjects, we see that age and dataset information are encoded simultaneously in the embedding space in different directions. Furthermore, we can pick vectors corresponding to an old and a young subject and generate their respective connectivity networks and study the differences in spatial configuration as well as frequency contents.
·Figure 1: M-DyNeMo finds uncoupled power and FC dynamics.
·Figure 2: Embedding vectors learns different sources of variation and can generate connectomes under different conditions.
Conclusions:
We propose two extensions to current generative models for estimating dynamic FC. The first method reveals novel insights into the evoked network response to task and ongoing activity that previous methods fail to capture, challenging the assumption that power and FC share the same dynamics.
The second method allows the discovery of different sources of variation and the more accurate inference of variability in FC, which can be useful in normative modelling and finding individualised biomarkers.
The second method allows the discovery of different sources of variation and the more accurate inference of variability in FC, which can be useful in normative modelling and finding individualised biomarkers.
Modeling and Analysis Methods:
Bayesian Modeling
Connectivity (eg. functional, effective, structural) 1
EEG/MEG Modeling and Analysis
Methods Development 2
Keywords:
Computational Neuroscience
Data analysis
Machine Learning
MEG
Modeling
Multivariate
Open-Source Code
Open-Source Software
Statistical Methods
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
Huang, R. (2024a). Modelling variability in functional brain networks using embeddings. bioRxiv, 2024-01.
Huang, R. (2024b). Evidence for transient, uncoupled power and functional connectivity dynamics. bioRxiv, 2024-08.