Multi-Tasking and Choice of Training Data Influencing Parietal ERP Expression and Single-Trial Detection—Relevance for Neuroscience and Clinical Applications

Event-related potentials (ERPs) are often used in brain-computer interfaces (BCIs) for communication or system control for enhancing or regaining control for motor-disabled persons. Especially results from single-trial EEG classification approaches for BCIs support correlations between single-trial ERP detection performance and ERP expression. Hence, BCIs can be considered as a paradigm shift contributing to new methods with strong influence on both neuroscience and clinical applications. Here, we investigate the relevance of the choice of training data and classifier transfer for the interpretability of results from single-trial ERP detection. In our experiments, subjects performed a visual-motor oddball task with motor-task relevant infrequent (targets), motor-task irrelevant infrequent (deviants), and motor-task irrelevant frequent (standards) stimuli. Under dual-task condition, a secondary senso-motor task was performed, compared to the simple-task condition. For evaluation, average ERP analysis and single-trial detection analysis with different numbers of electrodes were performed. Further, classifier transfer was investigated between simple and dual task. Parietal positive ERPs evoked by target stimuli (but not by deviants) were expressed stronger under dual-task condition, which is discussed as an increase of task emphasis and brain processes involved in task coordination and change of task set. Highest classification performance was found for targets irrespective whether all 62, 6 or 2 parietal electrodes were used. Further, higher detection performance of targets compared to standards was achieved under dual-task compared to simple-task condition in case of training on data from 2 parietal electrodes corresponding to results of ERP average analysis. Classifier transfer between tasks improves classification performance in case that training took place on more varying examples (from dual task). In summary, we showed that P300 and overlaying parietal positive ERPs can successfully be detected while subjects are performing additional ongoing motor activity. This supports single-trial detection of ERPs evoked by target events to, e.g., infer a patient's attentional state during therapeutic intervention.

Products GmbH) with reference at FCz. The subjects played a physical simulation of the BRIO R labyrinth and had to respond to infrequent target stimuli by pressing a buzzer. Deviant (infrequent non-target) stimuli were not presented to the subjects. The subjects were instructed to wait before responding to target stimuli until they had guided the ball into a safe corner. The 6 data sets of two subjects were merged into one data set for each subject. To find how the task reaction time influences the classification performance, the data was divided into 5 groups based on reaction time (RT in ms): RT 1400, RT 1600, RT 1800, RT 2000, and RT 7000.
The data for each group was resampled 10 times using bootstrapping technique by randomly choosing 60 training examples. To avoid the influence of training size, the number of training examples for target class (n = 60) was kept constant for each RT group. The remaining examples were used to test the classifier. Finally, we obtained 20 classification values for each RT group (10 values x 2 subjects for each RT group). As a performance metric, we used balanced accuracy (bACC).
To find an effect of task reaction time on the single-trial ERP detection, the data was analyzed by repeated measures ANOVA with reaction time (RT) as a within-subjects factor (5 levels: RT 1400, RT 1600, RT 1800, RT 2000, RT 7000). For pairwise comparisons, Bonferroni correction was applied. The observed ERP amplitude difference (Fig. S8), i.e., average ERP across two subjects between targets and standards was not statistically analyzed due to small sample sizes.
The classification performance of the positive parietal ERP complex was reduced by the delay of the response (see Fig. S7).The more the task response time was delayed, the more the classification performance of the positive parietal ERP complex was reduced. The intensity of the positive parietal ERP complex was reduced by the delay of the response (see Fig. S8). The more the task response time was delayed, the more the amplitude of the positive parietal ERP complex was reduced.  Figure S1. Average ERP of individual subjects for both task conditions (simple and dual task).  Figure S2. Statistical results on average ERP activity: comparison between stimulus types (A) and comparison between task conditions (B). Insignificant differences are marked in red.
All channels early window Parietal channels early window  Figure S4. Statistical results on single trial detection: comparison between channel types (62 channels, 6 and 2 parietal channels). Insignificant differences are marked in red.  Figure S5. Statistical results on single-trial detection: comparison between task types (simple and dual task) and transfer types (transfer case and no transfer case). Insignificant differences are marked in red. Figure S6. Statistical results on single-trial detection: comparison between window types (early and late window). Insignificant differences are marked in red. Further statistical values between task conditions and transfer conditions across three types of channels, both window types, and three different ERP detection types are reported: du vs. si → du: p < 0.001, du vs. du → si: p < 0.002, si vs. du → si: p < 0.001, si vs. si → du: p < 0.001.  Figure S7. Classification performance in the detection of the positive parietal ERP complex using machine learning. The mean of 20 balanced accuracy (bACC) for each RT group and standard error of mean is depicted. The classification performance of the positive parietal ERP complex was reduced by the delay of the response. The more the task reaction time was delayed, the smaller the classification performance of the positive parietal ERP complex became. Figure is lightly changed from Kim and Kirchner (2012).