Hierarchical Hidden Markov Model for Online Decoding in Brain-Computer Interface
- 26 August 2024
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1466-1470
Abstract
This paper presents the use of a hierarchical hidden Markov model (H2M2) for decoding brain signals from a tetraplegic patient. The H2M2 is a dynamic classifier used in this study to decode the user's motor intentions. A H2M2 consists of multiple hidden Markov models (HMM), with states organised into production states, which emit observations, and internal states, which activate sub-HMMs. The hierarchical organization of the decoder can be parameterized to represent the state to classify. A two layers and a three layers H2M2 architectures are compared with a classical HMM to decode up to seven states from ElectroCorticoGraphy (ECoG). Results show the H2M2 outperforms the HMM in term of classification performances and latency. The H2M2 presents also more stable results.Keywords
Funding Information
- European Commission (EIC 2021-TransitionChallenges-01-01)