Predicting individual traits from models of brain dynamics using the Fisher kernel
Poster No:
1096
Submission Type:
Abstract Submission
Authors:
Christine Ahrends1, Mark Woolrich2, Diego Vidaurre3
Institutions:
1University of Oxford, Oxford, United Kingdom, 2University of Oxford, Oxford, Oxon, 3Aarhus University, Aarhus, Denmark
First Author:
Co-Author(s):
Introduction:
Observing a person's behaviour over time is how we understand the individual's personality, cognitive traits, or psychiatric condition. The same should apply at the brain level, where we may be able to gain crucial insights by observing the patterns in which brain activity unfolds over time, i.e., brain dynamics. One way of describing brain dynamics are state-space models, which allow capturing recurring patterns of activity and functional connectivity (FC) across the whole brain (Liegeois et al., 2019, Vidaurre et al., 2021). However, it is still unclear how best to use this spatiotemporal level of description to predict individual traits from brain signals. We here propose the Fisher kernel (Jaakkola et al., 1999, Jakkola & Haussler, 1998), a mathematically principled and computationally efficient approach to predict phenotypes from a Hidden Markov Model (HMM) as a model of brain dynamics (Vidaurre et al., 2017). Critically, the Fisher kernel takes the complex relationships between the model parameters into account by preserving the structure of the underlying HMM.
Methods:
We started with the concatenated resting-state fMRI timeseries of 1,001 subjects from the Human Connectome Project (HCP) (van Essen et al., 2013). We estimated a Hidden Markov Model (HMM), which is a state-space model of time-varying amplitude and FC, first on the group- and then on the individual subject-level. The parameters of this model lie on a Riemannian manifold. In the Fisher kernel, the parameters are mapped into the gradient space, which is a tangent space to the Riemannian manifold. We compare the Fisher kernel to kernels which ignore the structure of the parameter space (naïve kernels), and to methods based on time-averaged FC. We then used these kernels or feature matrices to predict the behavioural variables using (kernel) ridge regression. We repeated the prediction in 35 different demographic and cognitive items using 100 randomised iterations of 10-fold nested cross validation (CV) to assess reliability.
·Figure 1. Overview of the HMM-Fisher kernel approach for predicting individual traits from models of brain dynamics.
Results:
Among the kernels constructed from HMMs, the Fisher kernel had the highest prediction accuracy on average across the range of behavioural variables and CV folds and iterations. This indicates a positive effect of using a tangent space embedding rather than incorrectly treating the HMM parameters as Euclidean. The Fisher kernel also has a higher prediction accuracy than several methods using time-averaged FC, but it is outperformed by time-averaged methods that work in tangent space (analogous to the effect seen for the time-varying methods). The Fisher kernel and the time-averaged FC tangent space method were also overall the most reliable methods: They never produced excessive errors (surpassing the original target variable by orders of magnitude), and robustly predicted at similar accuracies across the different CV folds and iterations.
Conclusions:
We here aimed to establish an approach that allows leveraging a rich description of the patterns in which brain activity unfolds over time to predict individual traits. We showed that the HMM-Fisher kernel approach accurately and reliably predicts traits from brain dynamics models trained on neuroimaging data. While we here focussed on fMRI, the method can also be applied to other modalities like MEG or EEG. It can also be straightforwardly implemented in any kernel-based prediction model or classifier and combined with other modalities (e.g., static or structural brain measures). This will allow gaining crucial insights into cognition and behaviour from how brain function changes over time and may have potential benefits for a variety of clinical goals, e.g., to diagnose or predict individual patients' outcomes, find biomarkers, or to deepen our understanding of changes in the brain related to treatment responses like drugs or non-pharmacological therapies.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling
Methods Development 2
Task-Independent and Resting-State Analysis
Keywords:
Data analysis
Informatics
Machine Learning
Modeling
Open-Source Code
Other - Dynamics
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:
For human MRI, what field strength scanner do you use?
Provide references using APA citation style.
2. Vidaurre, D., Llera, A., Smith, S. M., & Woolrich, M. W. (2021). Behavioural relevance of spontaneous, transient brain network interactions in fMRI. Neuroimage, 229, 117713.
3. Jaakkola, T., Diekhans, M., & Haussler, D. (1999). Using the Fisher kernel method to detect remote protein homologies. Proc Int Conf Intell Syst Mol Biol, 149-158.
4. Jaakkola, T., & Haussler, D. (1998). Exploiting Generative Models in Discriminative Classifiers. NIPS, 11, 487-493.
5. Vidaurre, D., Smith, S. M., & Woolrich, M. W. (2017). Brain network dynamics are hierarchically organized in time. Proc Natl Acad Sci U S A, 114(48), 12827-12832.
6. Van Essen, D. C. et al. (2013). The WU-Minn Human Connectome Project: an overview. Neuroimage, 80, 62-79.