Class-Aware Hidden Markov Model for functional connectivity estimation and classification
Poster No:
1417
Submission Type:
Abstract Submission
Authors:
Chendi Han1, Pavithran Pattiam Giriprakash1, Rajesh Nandy2, Dietmar Cordes1
Institutions:
1Cleveland Clinic, Las Vegas, NV, 2University of North Texas Health Science Center, Fort Worth, TX
First Author:
Co-Author(s):
Introduction:
Several approaches have been adopted to characterize the temporal dynamics of resting-state fMRI connectivity. Compared to sliding window methods (Allen, 2014; Cai, 2017), Hidden Markov Models (HMM) are suitable for rapid state transitions, with the covariance matrix directly measuring brain connectivity (Vidaurre, 2017; Zhang, 2019). In this paper, we propose a framework to classify fMRI data based on HMM. Compared with the two-step approach, feature extraction and subsequent classification based on selected features (Ji, 2012; Zhao, 2022; Jin, 2023; Canal-Garcia, 2024), our proposed method using the joint probability directly to computes the posterior probability, which is more efficient and maintain the interpretability (Quattoni, 2007; Sutton, 2012). We evaluate the classification accuracy using real fMRI datasets.
Methods:
Consider fMRI data labeled as X∈R^{N*p*T}, indicating N subjects with length T and p components. Each subject has a ground truth class label among C classes. We assume fMRI data in each class follow a time-independent Hidden Markov Model (HMM). Specifically, the model has K^{(c)} hidden states, with transition probability matrix A^{(c)}∈R^{K^{(c)}*K^{(c)}}, and the emission probability follows a Gaussian distribution with mean μ^{(c)}∈R^{K^{(c)}*p} and covariance σ^{(c)}∈R^{K^{(c)}*p*p}. The covariance matrix is regarded as the dynamic functional connectivity (Vidaurre, 2017; Zhang, 2019). During parameter estimation, we assume that K^{(c)}=3 for all classes, while other parameters θ={A,μ,σ} are treated as optimization parameters. For each class, the Expectation-Maximization (EM) algorithm can be used to maximize the joint probability Ψ, as shown in Figure 1(a). One potential problem is that performing the EM algorithm separately for each class may not generate good class separation. We implement additional gradient descent to minimize the Cross Entropy Loss (CEL) between the posterior probability and the ground truth. This method is also called Hidden Conditional Random Field (HCRF) (Quattoni, 2007; Sutton, 2012), as shown in Figure 1(b). Suppose we have a new time series X_test with unknown class labels. To assign these labels, we use an unsupervised classifier based on an HMM and a supervised classifier based on an HCRF.
Results:
The real fMRI data containing 292 subjects with 74 Cognitively Normal (CN), 119 with Mild Cognitive Impairment (MCI), and 99 with Alzheimer's Disease (AD). Resting-state fMRI data were acquired with 3T, TR=3000 ms, TE=30 ms, flip angle=90°, FoV=220 mm, slice thickness=3.4 mm, EPI factor=64, echo spacing=0.72 ms. After group independent component analysis, the time series per subject has size T=135 and p=54. The data is split into training, evaluation, and testing sets in a 7/2/1 ratio. Figure 2 (a-c) shows the transition probabilities, means, and covariance matrices from class-level EM results. For the supervised training, we find that data augmentation in the time domain and Lasso regularization could help to reduce CEL in the validation dataset (Zhang, 2019). The whole process is repeated 100 times. Figure 2 (d) shows the performance of different methods. Our proposed HCRF shows the highest performance, with classification accuracy close to 50%.
Conclusions:
The key findings of this study are that one-step posterior probability can be used for disease classification, and that supervised HCRF shows better performance compared with unsupervised HMM.
This study was funded by NIH-R01AG071566-02 and NIH-P20GM109025-08.
This study was funded by NIH-R01AG071566-02 and NIH-P20GM109025-08.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling 1
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Data analysis
FUNCTIONAL MRI
Machine Learning
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?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
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?
Which processing packages did you use for your study?
Provide references using APA citation style.
Cai, B., et al. (2017). Estimation of dynamic sparse connectivity patterns from resting state fMRI, IEEE transactions on medical imaging, 37(5), p.1224-1234.
Vidaurre, D., et al. (2017). Brain network dynamics are hierarchically organized in time. Proceedings of the National Academy of Sciences, vol. 114, no. 48: p.12827-12832.
Zhang, G., et al. (2019). Estimating dynamic functional brain connectivity with a sparse hidden Markov model. IEEE transactions on medical imaging, 39(2): p. 488-498.
Ji, J., et al. (2021). Convolutional neural network with sparse strategies to classify dynamic functional connectivity. IEEE Journal of Biomedical and Health Informatics, 26(3), p.1219-1228.
Zhao, C., et al. (2022). Abnormal characterization of dynamic functional connectivity in Alzheimer's disease. Neural regeneration research, 17(9), pp.2014-2021.
Jin, H., et al. (2023). Dynamic functional connectivity MEG features of Alzheimer’s disease. NeuroImage, 281, p.120358.
Canal-Garcia, A., et al. (2024). Dynamic multilayer functional connectivity detects preclinical and clinical Alzheimer’s disease. Cerebral Cortex, 34(2), p.542.
Quattoni, A., et al. (2007). Hidden-state conditional random fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, 29(10), p.1848-1852.
Sutton, C. and McCallum, A. (2012). An introduction to conditional random fields. Foundations and Trends® in Machine Learning, 4(4), p.267-373.