2. Biosignals Processing

Objectives:

Reduce the negative effects of EEG non-stationary, changes in electrode setup, etc.;
Towards one-time calibration in self-paced P300-based BCIs;
Reduce the negative effects of variability between subjects (e.g. universal models in automatic sleep stage classification (ASSC).

One-time calibration for self-paced P300-based BCIs

Achievements:

A new method was proposed allowing the control of self-paced P300-based BCIs using one-time calibration. The method combines distance threshold adjustment (DTA) to dynamically adapt the separation boundary between target and non-control classes, and semi-supervised self-training approach (SSST) for automatic recalibration using unlabeled samples collected during the online usage. Classification accuracies were similiar to those obtained if calibration and operation would be made in the same session.

Fig. SSST + DTA classification architecture.

Automatic sleep stage classification (ASSC)

Achievements:

A systematic analysis and comparison of methods for preprocessing, feature extraction, feature transformation and normalization, feature selection, and classification methods have been made, providing usefull information for selecting methods;
Domain adaptation methods have been researched to tackle the problem of distribution differences between training and testing and the problem of database bias. Import Vector Machine based on KLR has been researched. In comparison with other methods (e.g. SVM) IVM-KLR showed the following advantages: exact posterior probability can be computed; and it is straightforward a multi-class extension formulation of the method;
An ASSC method consisting of a two-step classifier was developed. In the first step, epochs are classified using support vector machines (SVMs) spread into different nodes of a decision tree. In the post-processing step, the epochs suspected of misclassification (dubious classification) are tagged, and a new classification is suggested. Identification and correction are based on the AASM rules, and on misclassifications most commonly found/reported in ASSC literature.

Fig. (top) Diagram of dubious range detection and dubious range correction; (bottom) Hypnogram with dubious epochs

Publications

S. Khalighi, B. Ribeiro, U. Nunes, Importance weighted import vector machine for unsupervised domain adaptation IEEE Transactions on Cybernetics, 2016.
S. Khalighi, T. Sousa, J. Moutinho Santos, U. Nunes, ISRUC-Sleep: a comprehensive public dataset for sleep researchers, Computer Methods and Programs in Biomedicine, Esevier, 2015, doi:10.1016/j.cmpb.2015.10.013;
Teresa Sousa, Aniana Cruz, S. Khalighi, G. Pires, U. Nunes, A two-step automatic sleep stage classification method with dubious range detection, Computers in Biology and Medicine, Elsevier, 2015, doi:10.1016/j.compbiomed.2015.01.017 ;
S. Khalighi, T. Sousa, G. Pires, U. Nunes, Automatic sleep staging: A computer assisted approach for optimal combination of features and polysomnographic channels, Expert Systems with Applications, Elsevier, 2013, doi:10.1016/j.eswa.2013.06.023 ;
J. Figueiredo, G. Pires, U. Nunes, “Self-paced control of a P300-based BCI-speller using one-time calibration: performance analysis”, ( to be submitted to int. journal)

Datasets and code

PhD Theses

MSc and BSc Theses

Joana Figueiredo, Research of adaptive classification techniques towards one-time calibration in brain-computer interfaces, MSc dissertation (supervisors: U. Nunes and G. Pires), FCTUC, Sept. 2014;