Using HMM to Model Neural Dynamics and Decode Useful Signals for Neuroprosthetic Control

A central aim of modern neuroscience is to unravel how the various brain areas coordinate their activity across time and how they control behaviour. In this chapter, we first provide the rationale for the application of mathematical methods that analyse neural signals with a population approach. Among these techniques, Hidden Markov Models (HMMs) have been extensively applied. We then presented the key concepts of HMMs and the main problems during their application. Using neural data recorded from three distinct sectors in the monkey parietal cortex during an arm movement task, we show how HMMs can be used both to define neural dynamics and the flow of information at the network level, but also to decode relevant task parameters from neural activity in order to control finite state brain–machine interfaces (BMIs). We found that the HMM state transitions were in accordance with a functional gradient in the parietal cortex and that they carry reliable information about target position and behavioural phases. Finally, we provide a brief review of the literature on HMMs in the field and an overview of other approaches that can model time series.