Class-Aware Hidden Markov Model for simultaneous functional connectivity estimation and classification

Motivation: Hidden Markov Models (HMM) are widely used for modeling resting-state fMRI data. However, current classification approaches usually involve additional parameter extraction steps, which make the model complicated and reduce sensitivity. Goal(s): Our goal is to develop a one-step classification method based on HMMs that is more efficient, accurate, and maintains interpretability. Approach: We utilize Hidden Conditional Random Fields (HCRF), combining HMM and discriminative learning into a unified model that eliminates the need for separate parameter extraction. Results: In both simulated and real data, the one-step model outperforms the traditional two-step model. Supervised learning could further improve classification accuracy. Impact: For fMRI classification problems, the one-step Class-Aware HMM is simpler and more accurate compared to two-step classification while maintaining model interpretability. This could help in understanding brain connectivity and disease diagnosis.