Rebeca Corralejo

Adaptive semi-supervised classification to reduce intersession non-stationarity in multiclass motor imagery-based brain-computer interfaces

The intersession non-stationarity in electroencephalogram (EEG) data is a major issue to robust operation of brain-computer interfaces (BCIs). The aim of this paper is to propose a semi-supervised classification algorithm whereby the model is gradually enhanced with unlabeled data collected online. Additionally, a processing stage is introduced before classification to adaptively reduce the small fluctuations between the features from training and evaluation sessions. The key element of the classification algorithm is an optimized version of kernel discriminant analysis called spectral regression kernel discriminant analysis (SRKDA) in order to meet the low computational cost requirement for online BCI applications. Four different approaches, SRKDA and sequential updating semi-supervised SRKDA (SUSS-SRKDA) with or without adaptive processing stage are considered to quantify the advantages of semi-supervised learning and adaptive stage. The session-to-session performance for each of them is evaluated on the multiclass problem (four motor imagery tasks: the imagination of movement of the left hand, right hand, both feet, and tongue) posed in the BCI Competition IV dataset 2a. The results agree with previous studies reporting semi-supervised learning enhances the adaptability of BCIs to non-stationary EEG data. Moreover, we show that reducing the inter-session non-stationarity before classification further boosts its performance. The classification method combining adaptive processing and semi-supervised learning is found to yield the highest session-to session transfer results presented so far for this multiclass dataset: accuracy (77%) and Cohen?s kappa coefficient (0.70). Thus, the proposed methodology could be of great interest for real-life BCIs.

Feature selection using a genetic algorithm in a motor imagery-based Brain Computer Interface

This study performed an analysis of several feature extraction methods and a genetic algorithm applied to a motor imagery-based Brain Computer Interface (BCI) system. Several features can be extracted from EEG signals to be used for classification in BCIs. However, it is necessary to select a small group of relevant features because the use of irrelevant features deteriorates the performance of the classifier. This study proposes a genetic algorithm (GA) as feature selection method. It was applied to the dataset IIb of the BCI Competition IV achieving a kappa coefficient of 0.613. The use of a GA improves the classification results using extracted features separately (kappa coefficient of 0.336) and the winner competition results (kappa coefficient of 0.600). These preliminary results demonstrated that the proposed methodology could be useful to control motor imagery-based BCI applications.

Adaptive Stacked Generalization for Multiclass Motor Imagery-Based Brain Computer Interfaces

Practical motor imagery-based brain computer interface (MI-BCI) applications are limited by the difficult to decode brain signals in a reliable way. In this paper, we propose a processing framework to address non-stationarity, as well as handle spectral, temporal, and spatial characteristics associated with execution of motor tasks. Stacked generalization is used to exploit the power of classifier ensembles for combining information coming from multiple sources and reducing the existing uncertainty in EEG signals. The outputs of several regularized linear discriminant analysis (RLDA) models are combined to account for temporal, spatial, and spectral information. The resultant algorithm is called stacked RLDA (SRLDA). Additionally, an adaptive processing stage is introduced before classification to reduce the harmful effect of intersession non-stationarity. The benefits of the proposed method are evaluated on the BCI Competition IV dataset 2a. We demonstrate its effectiveness in binary and multiclass settings with four different motor imagery tasks: left-hand, right-hand, both feet, and tongue movements. The results show that adaptive SRLDA outperforms the winner of the competition and other approaches tested on this multiclass dataset.

Analysis of neural dynamics in mild cognitive impairment and Alzheimer's disease using wavelet turbulence

OBJECTIVE Current diagnostic guidelines encourage further research for the development of novel Alzheimers disease (AD) biomarkers, especially in its prodromal form (i.e. mild cognitive impairment, MCI). Magnetoencephalography (MEG) can provide essential information about AD brain dynamics; however, only a few studies have addressed the characterization of MEG in incipient AD. APPROACH We analyzed MEG rhythms from 36 AD patients, 18 MCI subjects and 27 controls, introducing a new wavelet-based parameter to quantify their dynamical properties: the wavelet turbulence. MAIN RESULTS Our results suggest that AD progression elicits statistically significant regional-dependent patterns of abnormalities in the neural activity (p < 0.05), including a progressive loss of irregularity, variability, symmetry and Gaussianity. Furthermore, the highest accuracies to discriminate AD and MCI subjects from controls were 79.4% and 68.9%, whereas, in the three-class setting, the accuracy reached 67.9%. SIGNIFICANCE Our findings provide an original description of several dynamical properties of neural activity in early AD and offer preliminary evidence that the proposed methodology is a promising tool for assessing brain changes at different stages of dementia.

Assessment of Neurofeedback Training by means of Motor Imagery based-BCI for Cognitive Rehabilitation

The age-related impairment is an increasing problem due to the aging suffered by the population, especially in developed countries. It is usual to use electroencephalogram (EEG)-based Brain Computer Interface (BCI) systems by means of the signal in order to assist and to improve the quality of life of people with disabilities. However, a parallel research line addresses the problem by the use of BCI systems as a way to train cognitive areas to achieve a deceleration of cognitive impairment or even an improvement. In this regard, a neurofeedback training (NFT) tool using motor imagery-based BCI, was developed. Training consists on imagery motor exercises combined with memory and logical relation tasks. In order to assess the effectiveness of the application 40 subjects, older than 59 years old, took part in this study. Our NFT application was tested by 20 subjects and their scores of a neuropsychological test were compared with the remaining 20 subjects who did not perform the NFT. Results show a significant improvement of three cognitive features after performing the NFT: visual perception, expressive speech, and immediate memory. Therefore, evidences show that the performance of a NFT tool based on motor imagery tasks could be a positive activity for slow down the aging effects.

Ensemble learning for classification of motor imagery tasks in multiclass brain computer interfaces

The difficulty to decode brain signals in a reliable way limits practical motor imagery-based brain computer interface (MI-BCI) applications. The aim of this paper is to propose a classification framework that handle spectral, temporal, and spatial characteristics associated with execution of motor imagery tasks, as well as the temporal variability in EEG data. An ensemble learning approach such as stacked generalization is used to combine information coming from multiple sources. The session-to-session performance of the proposed classifier ensemble is evaluated on a multiclass problem posed in the BCI Competition IV dataset 2a. The results yields a higher mean kappa of 0.66 compared to 0.62 from the baseline linear discriminant analysis (LDA). Also, our approach outperforms the winner of the BCI Competition IV dataset 2a and other studies reported in BCI literature.

Adaptive Classification Framework for Multiclass Motor Imagery-Based BCI

The non-stationary nature of electroencephalo- gram (EEG) is a major issue to robust operation of brain- computer interfaces (BCIs). The objective of this paper is to propose an adaptive classification framework whereby a processing stage is introduced before classification to address non-stationarity in EEG classification. Features extracted from EEG signals are adaptively processed before classification to reduce the small fluctuations between calibration and evalua- tion data. In this way, static classifiers can be employed in non- stationary environments without additional changes. The session-to-session performance of the proposed adaptive ap- proach is evaluated on a multiclass problem posed in the BCI Competition IV dataset 2a. A probabilistic generative model was used as a classification algorithm. The results yields a significantly higher mean kappa of 0.62 compared to 0.58 from the baseline probabilistic generative model without adaptive processing. Also, the proposed approach outperforms the winner of the BCI Competition IV dataset 2a. These results suggest a promising approach separating adaptation-related tasks and classification.

Analytic common spatial pattern and adaptive classification for multiclass motor imagery-based BCI

This paper focuses on the classification of motor imagery tasks from electroencephalogram (EEG) for brain computer interfaces (BCI). A new processing algorithm based on filter bank common spatial pattern (FBCSP) is presented. Analytic common spatial pattern (ACSP) and adaptive classification are introduced to investigate whether they can improve the performance. Four versions of FBCSP, namely, common spatial pattern (CSP) and ACSP with static or adaptive classification are studied. The session-to-session performances of the proposed approaches are evaluated on a 4-class problem posed in the BCI Competition IV dataset 2a. Our results demonstrate the effectiveness of the proposed methods in comparison to the winner of the BCI Competition IV Dataset 2a as well as other more recent studies using this dataset. Adaptive classification yields a higher kappa value of 0.61 compared to 0.57 for multiclass FBCSP algorithm. ACSP further improves the performance achieving a mean kappa of 0.63.

Neurofeedback Training with a Motor Imagery-Based BCI Improves Neurocognitive Functions in Elderly People

In recent years, brain-computer interfaces (BCIs) have become not only a tool to provide communication and control for people with disabilities, but also a way to rehabilitate some motor or cognitive functions. Brain plasticity can help restore normal brain functions by inducing brain activity. In fact, voluntary event-related desynchronization (ERD) in upper alpha and beta electroencephalogram (EEG) activity bands is associated with different neurocognitive functions. In this regard, neurofeedback training (NFT) has shown to be a suitable way to control one’s own brain activity. Furthermore, new evidence in recent studies showed NFT could lead to microstructural changes in white and grey matter. In our novel study, NFT qualities were applied to aging-related effects. We hypothesized that a NFT by means of motor imagery-based BCI (MI-BCI) could affect different cognitive functions in elderly people. To assess the effectiveness of this application, we studied 63 subjects, all above 60 years old. The subjects were divided into a control group (32 subjects) and a NFT group (31 subjects). To validate the effectiveness of the NFT using MI-BCI, variations in the scores of neuropsychological tests (Luria tests) were measured and analyzed. Results showed significant improvements (p < 0.05) in the NFT group, after only five NFT sessions, in four cognitive functions: visuospatial; oral language; memory; and intellectual. These results further support the association between NFT and the enhancement of cognitive performance. Findings showed the potential usefulness of NFT using MI-BCI. Therefore, this approach could lead to new means to help elderly people.

Assessment of an Assistive P300-Based Brain Computer Interface by Users with Severe Disabilities

The present study aims at assessing an assistive P300-based BCI tool for managing electronic devices at home. Fifteen subjects with motor and cognitive disabilities participated in the study. The assistive tool was designed to be simple and easy to interact with users. It allows managing 113 control commands from 8 different devices. Although most of the participants also showed cognitive impairments, nine out of the fifteen participants were able to properly manage the assistive BCI application with accuracy higher than 80%. Moreover, five out of them achieved accuracies higher than 95%. Maximum information transfer rate (ITR) values of 14.41 bits/min were reached. Hence, P300-based BCIs could be suitable for developing new control interfaces fulfilling the main needs of disabled people, such as comfort, communication and entertainment. Our results suggest that the degree of motor or cognitive disability is not a relevant issue in order to suitably operate the assistive BCI application.