David Steyrl

Random forests in non-invasive sensorimotor rhythm brain-computer interfaces: a practical and convenient non-linear classifier

Abstract There is general agreement in the brain-computer interface (BCI) community that although non-linear classifiers can provide better results in some cases, linear classifiers are preferable. Particularly, as non-linear classifiers often involve a number of parameters that must be carefully chosen. However, new non-linear classifiers were developed over the last decade. One of them is the random forest (RF) classifier. Although popular in other fields of science, RFs are not common in BCI research. In this work, we address three open questions regarding RFs in sensorimotor rhythm (SMR) BCIs: parametrization, online applicability, and performance compared to regularized linear discriminant analysis (LDA). We found that the performance of RF is constant over a large range of parameter values. We demonstrate – for the first time – that RFs are applicable online in SMR-BCIs. Further, we show in an offline BCI simulation that RFs statistically significantly outperform regularized LDA by about 3%. These results confirm that RFs are practical and convenient non-linear classifiers for SMR-BCIs. Taking into account further properties of RFs, such as independence from feature distributions, maximum margin behavior, multiclass and advanced data mining capabilities, we argue that RFs should be taken into consideration for future BCIs.

Short time sports exercise boosts motor imagery patterns: implications of mental practice in rehabilitation programs

Motor imagery (MI) is a commonly used paradigm for the study of motor learning or cognitive aspects of action control. The rationale for using MI training to promote the relearning of motor function arises from research on the functional correlates that MI shares with the execution of physical movements. While most of the previous studies investigating MI were based on simple movements in the present study a more attractive mental practice was used to investigate cortical activation during MI. We measured cerebral responses with functional magnetic resonance imaging (fMRI) in twenty three healthy volunteers as they imagined playing soccer or tennis before and after a short physical sports exercise. Our results demonstrated that only 10 min of training are enough to boost MI patterns in motor related brain regions including premotor cortex and supplementary motor area (SMA) but also fronto-parietal and subcortical structures. This supports previous findings that MI has beneficial effects especially in combination with motor execution when used in motor rehabilitation or motor learning processes. We conclude that sports MI combined with an interactive game environment could be a promising additional tool in future rehabilitation programs aiming to improve upper or lower limb functions or support neuroplasticity.

A co-adaptive sensory motor rhythms Brain-Computer Interface based on common spatial patterns and Random Forest

Sensorimotor rhythm (SMR) based Brain-Computer Interfaces (BCI) typically require lengthy user training. This can be exhausting and fatiguing for the user as data collection may be monotonous and typically without any feedback for user motivation. Hence new ways to reduce user training and improve performance are needed. We recently introduced a two class motor imagery BCI system which continuously adapted with increasing run-time to the brain patterns of the user. The system was designed to provide visual feedback to the user after just five minutes. The aim of the current work was to improve user-specific online adaptation, which was expected to lead to higher performances. To maximize SMR discrimination, the method of filter-bank common spatial patterns (fbCSP) and Random Forest (RF) classifier were combined. In a supporting online study, all volunteers performed significantly better than chance. Overall peak accuracy of 88.6 ± 6.1 (SD) % was reached, which significantly exceeded the performance of our previous system by 13%. Therefore, we consider this system the next step towards fully auto-calibrating motor imagery BCIs.

Cooperation in mind: Motor imagery of joint and single actions is represented in different brain areas.

In this study brain activity during motor imagery (MI) of joint actions, compared to single actions and rest conditions, was investigated using functional magnetic resonance imaging (fMRI). To the best of our knowledge, this is the first neuroimaging study which directly investigated the neural correlates of joint action motor imagery. Twenty-one healthy participants imagined three different motor tasks (dancing, carrying a box, wiping). Each imagery task was performed at two kinds: alone (single action MI) or with a partner (joint action MI). We hypothesized that to imagine a cooperative task would lead to a stronger cortical activation in motor related areas due to a higher vividness and intensification of the imagery. This would be elicited by the integration of the action simulation of the virtual partner to ones own action. Comparing the joint action and the single action condition with the rest condition, we found significant activation in the precentral gyrus and precuneus respectively. Furthermore the joint action MI showed higher activation patterns in the premotor cortex (inferior and middle frontal gyrus) compared to the single action MI. The imagery of a more vivid and engaging task, like our joint action imagery, could improve rehabilitation processes since a more distributed brain activity is found. Furthermore, the joint action imagery compared to single action imagery might be an appropriate BCI task due to its clear spatial distinction of activation.

Single trial classification of fNIRS-based brain-computer interface mental arithmetic data: A comparison between different classifiers

Functional near infrared spectroscopy (fNIRS) is an emerging technique for the in-vivo assessment of functional activity of the cerebral cortex as well as in the field of brain-computer-interface (BCI) research. A common challenge for the utilization of fNIRS for BCIs is a stable and reliable single trial classification of the recorded spatio-temporal hemodynamic patterns. Many different classification methods are available, but up to now, not more than two different classifiers were evaluated and compared on one data set. In this work, we overcome this issue by comparing five different classification methods on mental arithmetic fNIRS data: linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), support vector machines (SVM), analytic shrinkage regularized LDA (sLDA), and analytic shrinkage regularized QDA (sQDA). Depending on the used method and feature type (oxy-Hb or deoxy-Hb), achieved classification results vary between 56.1 % (deoxy-Hb/QDA) and 86.6% (oxy-Hb/SVM). We demonstrated that regularized classifiers perform significantly better than non-regularized ones. Considering simplicity and computational effort, we recommend the use of sLDA for fNIRS-based BCIs.

Random Forests for Feature Selection in Non-invasive Brain-Computer Interfacing

The aim of the present study was to evaluate the usefulness of the Random Forest (RF) machine learning technique for identifying most significant frequency components in electroencephalogram (EEG) recordings in order to operate a brain computer interface (BCI). EEG recorded from ten able-bodied individuals during sustained left hand, right hand and feet motor imagery was analyzed offline and BCI simulations were computed. The results show that RF, within seconds, identified oscillatory components that allowed generating robust and stable BCI control signals. Hence, RF is a useful tool for interactive machine learning and data mining in the context of BCI.

Reference layer adaptive filtering (RLAF) for EEG artifact reduction in simultaneous EEG-fMRI

OBJECTIVE Simultaneous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) combines advantages of both methods, namely high temporal resolution of EEG and high spatial resolution of fMRI. However, EEG quality is limited due to severe artifacts caused by fMRI scanners. APPROACH To improve EEG data quality substantially, we introduce methods that use a reusable reference layer EEG cap prototype in combination with adaptive filtering. The first method, reference layer adaptive filtering (RLAF), uses adaptive filtering with reference layer artifact data to optimize artifact subtraction from EEG. In the second method, multi band reference layer adaptive filtering (MBRLAF), adaptive filtering is performed on bandwidth limited sub-bands of the EEG and the reference channels. MAIN RESULTS The results suggests that RLAF outperforms the baseline method, average artifact subtraction, in all settings and also its direct predecessor, reference layer artifact subtraction (RLAS), in lower (<35 Hz) frequency ranges. MBRLAF is computationally more demanding than RLAF, but highly effective in all EEG frequency ranges. Effectivity is determined by visual inspection, as well as root-mean-square voltage reduction and power reduction of EEG provided that physiological EEG components such as occipital EEG alpha power and visual evoked potentials (VEP) are preserved. We demonstrate that both, RLAF and MBRLAF, improve VEP quality. For that, we calculate the mean-squared-distance of single trial VEP to the mean VEP and estimate single trial VEP classification accuracies. We found that the average mean-squared-distance is lowest and the average classification accuracy is highest after MBLAF. RLAF was second best. SIGNIFICANCE In conclusion, the results suggests that RLAF and MBRLAF are potentially very effective in improving EEG quality of simultaneous EEG-fMRI. Highlights We present a new and reusable reference layer cap prototype for simultaneous EEG-fMRI We introduce new algorithms for reducing EEG artifacts due to simultaneous fMRI The algorithms combine a reference layer and adaptive filtering Several evaluation criteria suggest superior effectivity in terms of artifact reduction We demonstrate that physiological EEG components are preserved.

Reduction of EEG artifacts in simultaneous EEG-fMRI: Reference layer adaptive filtering (RLAF)

Although simultaneous measurement of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) is one of the most valuable methods for studying human brain activity non-invasively, it remains challenging to measure high quality EEG inside the MRI scanner. Recently, a new approach for minimizing residual MRI scanner artifacts in the EEG was presented: reference layer artifact subtraction (RLAS). Here, reference electrodes capture only the artifacts, which are subsequently subtracted from the measurement electrodes. With the present work we demonstrate that replacing the subtraction by adaptive filtering statistically significantly outperforms RLAS. Reference layer adaptive filtering (RLAF) attenuates the average artifact root-mean-square (RMS) voltage of the passive MRI scanner to 0.7 μV (-14.4 dB). RLAS achieves 0.78 μV (-13.5 dB). The combination of average artifact subtraction (AAS) and RLAF reduces the residual average gradient artifact RMS voltage to 2.3 μV (-49.2 dB). AAS alone achieves 5.7 μV (-39.0 dB). All measurements were conducted with an MRI phantom, as the reference layer cap available to us was a prototype.

Adaptive hybrid brain-computer interaction: Ask a trainer for assistance!

In applying mental imagery brain-computer interfaces (BCIs) to end users, training is a key part for novice users to get control. In general learning situations, it is an established concept that a trainer assists a trainee to improve his/her aptitude in certain skills. In this work, we want to evaluate whether we can apply this concept in the context of event-related desynchronization (ERD) based, adaptive, hybrid BCIs. Hence, in a first session we merged the features of a high aptitude BCI user, a trainer, and a novice user, the trainee, in a closed-loop BCI feedback task and automatically adapted the classifier over time. In a second session the trainees operated the system unassisted. Twelve healthy participants ran through this protocol. Along with the trainer, the trainees achieved a very high overall peak accuracy of 95.3 %. In the second session, where users operated the BCI unassisted, they still achieved a high overall peak accuracy of 83.6%. Ten of twelve first time BCI users successfully achieved significantly better than chance accuracy. Concluding, we can say that this trainer-trainee approach is very promising. Future research should investigate, whether this approach is superior to conventional training approaches. This trainer-trainee concept could have potential for future application of BCIs to end users.

Single Trial Motor Imagery Classification in EEG Measured During FMRI Image Acquisition - A First Glance.

Non-invasive electroencephalogram (EEG) based Brain-Computer Interfaces (BCIs), which rely on event related desynchronization (ERD), are often affected by large fluctuations of their accuracy. We want to overcome this drawback by using simultaneous EEG and functional magnetic imaging (fMRI). The question we are addressing in this work is if ERD is still classifiable in EEG on a single trial basis after the removement of fMRI related artefacts. In a first single participant recording we found the classical ERD distribution and were able to compute a leave-one-out-cross-validation (LOOCV) accuracy of 78%, which is significantly higher than chance level.