Using Hidden Markov Models to Classify Brain Activity from fMRI Data
Author Information
Author(s): Grace Huckins, Russell A. Poldrack
Primary Institution: Stanford University
Hypothesis
Can hidden Markov models effectively classify resting-state fMRI data based on brain dynamics?
Conclusion
Hidden Markov models can classify resting-state fMRI data with promising performance, particularly for within-subject classification.
Supporting Evidence
- Hidden Markov models achieved around 80% accuracy in classifying caffeinated versus uncaffeinated states.
- The transition matrix-only approach performed similarly to fully trained models, indicating the importance of dynamic features.
- Classification performance was validated on held-out data, showing consistent results with cross-validation.
- Family misidentifications were rare, suggesting the model's robustness against familial similarities.
Takeaway
This study shows that we can use special models to understand how our brains work when we're not doing anything, helping us tell if someone has had caffeine or not just by looking at their brain scans.
Methodology
The study used hidden Markov models to classify resting-state fMRI data based on transition probabilities among hidden states.
Potential Biases
Potential confounding effects from caffeine and head motion were not fully controlled.
Limitations
The study's findings may not generalize to all classification tasks, and the optimal number of hidden states for classification remains uncertain.
Participant Demographics
One male participant aged 45 years for the MyConnectome dataset; 196 participants from the Human Connectome Project.
Statistical Information
P-Value
p<0.001
Confidence Interval
95% confidence intervals over 100 runs of cross-validation.
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website