Study of stroke condition and hand dominance using a hidden Markov, multivariate autoregressive (HMM-mAR) network framework

To investigate the effects of stroke and hand dominance on muscle association patterns during reaching movements, we applied the hidden Markov model, multivariate autoregressive (HMM-mAR) framework to real sEMG recordings from healthy and stroke subjects performing reaching tasks. Statistical analysis is performed to construct subject- and group-level muscle connectivity networks. Associating structural features are extracted for subsequent classification of reaching movements. The HMM-mAR framework is shown to be able to consistently segments each reaching movement into the initial phase and the full-movement phase. The inferred muscle networks illustrate that healthy and stroke subjects use distinguishably different muscle synergies during the initial phase. The classification results further confirm that structural features extracted from the initial phase are useful in classifying subjects with differing stroke condition and handedness.