Correção de artefatos e teste de hipóteses baseados em modelos da resposta mismatch negativity em sinais de EEG para aplicações em tempo real

Abstract

Brain-Machine Interfaces use brain signals for real-time control of various assisitve systems, such as alternative communication devices, upper-limb prostheses, exoskeletons, wheelchairs, etc. These approaches face intrinsic challenges, such as the removal of artifacts to extract reliable features in single-trial, especially when the cortical activity is measured by electroencephalography (EEG). In addition to the artifacts, another problem is the cortical response, since it presents variability between subjects and over time. Some studies have also shown that there is a statistically significant difference in cortical response between healthy subjects and Amyotrophic Lateral Sclerosis patients. In this way, computational models compare different models and identify which one best explains the cortical signal by EEG. However, developing experiments able to conclusively discriminate models is not trivial. However, developing experiments able to conclusively discriminate models is not trivial. Therefore, the hypothesis of this research is that the combined use of real-time artifact correction and an optimized adaptive system can provide more accurate and faster responses to cortical sensory perception through EEG data in online and single-trial analyses. There are some approaches to online artifact correction. Still, none is being used as a “gold-standard”, and no research has been conducted to analyze and compare their respective effects by employing inference-based decision, that is, the comparison of mathematical models that aim to explain the cortical dynamics due to some external event. Therefore, in the first part of this research, we evaluated methods for artifact correction and the necessary adjustments to implement them in single-trial for online electroencephalographic (EEG) analysis. We investigate the following methods: Artifact Subspace Reconstruction (ASR), Fully Online and automated artifact Removal for brain-Computer interfacing (FORCe), Empirical Model Decomposition online (EMD), and Independent Component Analysis online (ICA). For assessment, we simulate online data processing using real data from an oddball auditory task. We compare the above methods with a classical offline data processing, in their ability (i) to reveal a significant Mismatch Negativity (MMN) response to auditory stimuli; (ii) to reveal the more subtle modulation of the MMN by contextual changes (namely, the predictability of the sound sequence) and (iii) to identify the cortical process modeling of sensory perception most likely to explain the MMN response. Our results show that ASR and EMD are both able to reveal MMN and its modulation by predictability, and even appear more sensitive than the offline analysis when comparing alternative models of perception underlying auditory evoked responses. In the second part of this research, we specifically explore cortical modeling. Some studies propose the implementation of adaptive designs, which make it possible to distinguish between models faster and more accurately. However, to date, no study has explored hypothesis testing (model comparison) considering sensory perception computational modeling (MMN) to optimize real-time experimental designs based on single-trial EEG signals. Our results with simulated data showed that the hypothesis testing was able to conclude in favor of the model that generated the data. Besides, adaptive design showed better results than the classic design. The results with real data showed variability in the cortical response between subjects and also in terms of the experimental block (predictability in the sound sequence). The adaptive design showed more results in favor of alternative models, while the classic design, in favor of the null model. The combination of real-time artifact correction and adaptive design proved to be feasible to identify the computational model that best explains the cortical response in single-trial analysis of the EEG signal. These results are important for ICM applications as they can identify whether the subject is able to use the system and also to investigate changes in his cortical response over time.