Estimating Overlapped Event-Related Response with EM algorithm
Presenting author:
The study of event-related potentials (ERP) has become increasingly popular in the last decades because it provides a continuous measure of processing between a stimulus and a response with an excellent temporal resolution. Still, with the rapid presentation of objects in this type of experimental paradigm, the waveforms are extremely entangled in the time domain, which causes that the use of deconvolution methods be slow or difficult to converge. The most relevant studies use select peaks or mean amplitudes of averaged evoked responses that may produce results that are not significant statistically.
The use of methods that assume variability in the response across subjects, like the mixed effect models, offers many improvements for EEG modeling due to its higher power to detect variability across subjects and groups. Still, its use in real data requires deconvolution methods for continuous EEG data, which have a high computational cost.
They are a few software that solves both problems: The overlapping neural responses from subsequent events and the use of robust models with high statistical power to detect variability. The unmixed toolbox expands the unfold Matlab toolbox for the Linear Mixed Model, and it is the only one that we know that uses Linear Mixed Models and overlap correction. However, this program's use for real EEG data is still complicated due to the optimizer speed and convergence, which take a lot of time.
We propose a software that uses overlap correction and a robust model for the solution of both problems. It is based on the EM algorithm for the estimator's calculus using a Linear Mixed Effect Model. This software is numerically stable with high precision and speed. It uses an excellent initial approximation that permits a quick convergence of the EM algorithm. The software was tested with simulation and was applied to real data. We use the data from The Barbados Nutrition Study (BNS), a longitudinal study on a Barbadian cohort with histories of moderate to severe protein-energy malnutrition (PEM) limited to the first year of life and a healthy comparison group. The results obtained show how our methods are capturing the differences between both groups, malnourished and control.
The use of methods that assume variability in the response across subjects, like the mixed effect models, offers many improvements for EEG modeling due to its higher power to detect variability across subjects and groups. Still, its use in real data requires deconvolution methods for continuous EEG data, which have a high computational cost.
They are a few software that solves both problems: The overlapping neural responses from subsequent events and the use of robust models with high statistical power to detect variability. The unmixed toolbox expands the unfold Matlab toolbox for the Linear Mixed Model, and it is the only one that we know that uses Linear Mixed Models and overlap correction. However, this program's use for real EEG data is still complicated due to the optimizer speed and convergence, which take a lot of time.
We propose a software that uses overlap correction and a robust model for the solution of both problems. It is based on the EM algorithm for the estimator's calculus using a Linear Mixed Effect Model. This software is numerically stable with high precision and speed. It uses an excellent initial approximation that permits a quick convergence of the EM algorithm. The software was tested with simulation and was applied to real data. We use the data from The Barbados Nutrition Study (BNS), a longitudinal study on a Barbadian cohort with histories of moderate to severe protein-energy malnutrition (PEM) limited to the first year of life and a healthy comparison group. The results obtained show how our methods are capturing the differences between both groups, malnourished and control.