Classification of Neural Correlates of Error Processing

Abstract

Introduction

Detection and classification of correlates of error processing is a key area of interest in neuroscience and has important practical implications for fields such as brain-computer interfaces and human-computer interaction. One approach to this problem is through the use of electroencephalography (EEG), which can detect the electrical potential produced by the brain when an error occurs [1]. In this study, we aim to measure and classify two distinct types of errors, namely execution errors and outcome errors [2]. Our primary research questions are: Can we differentiate between two distinct types of errors - execution errors and outcome errors - in the context of a complex motor task? Moreover, can we predict outcome errors before the actual outcome of the task?

Methods

To answer these questions, we employed a visuomotor rotation paradigm - a technique wherein visual feedback of a participant's motor action is systematically altered to induce sensory-motor errors in a group of participants (n=7) while their brain activity was recorded with EEG. Specifically, the participants were tasked with performing a reaching arm movement with a haptic robot, aiming to hit a designated target. Perturbations in the form of random rotations at varying angles (± 40°, ± 20°) were introduced during 20 % of the trials. The experiment was divided into three experimental blocks, each designed to induce a specific type of error. To investigate the feasibility of differentiating between different types of errors, we conducted four binary classifications using multiple neural network models based on convolutional and transformer architectures. These classifications included outcome error vs. no error, execution error vs. no error, outcome error vs. execution error, and pre-outcome error vs. no error.

Expected Results

Previous work suggests that such differentiation is possible in a cursor control video game paradigm [2], but it is unclear to what extent these results apply to more complex motor tasks which involve more noisy experimental settings. Based on preliminary analysis, we found that in a cross-subject classification paradigm, in which we train a general classifier from all subject data, accuracies in all binary classifications were not above the chance level. However, within-subject classification resulted in above-chance level accuracies for some models and subjects.

References

[1] R. Chavarriaga, A. Sobolewski, and J. del R. Millán, “Errare machinale est: the use of error-related potentials in brain-machine interfaces,” Frontiers in Neuroscience, vol. 8, p. 208, 2014. doi: 10.3389/fnins.2014.00208

[2] M. Spüler and C. Niethammer, “Error-related potentials during continuous feedback: using EEG to detect errors of different type and severity,” Frontiers in Human Neuroscience, vol. 9, p. 155, 2015. doi: 10.3389/fnhum.2015.00155