Decoding brain states using functional brain imaging techniques.

摘要

Non-invasive neuroimaging techniques provide safe methods for investigating the functionality of the brain. Functional near infrared spectroscopy (fNIRS) is a non invasive brain imaging method, which uses light in the near infrared range to measure the changes in concentration of cerebral hemoglobin. Electroencephalography (EEG) is a noninvasive brain imaging technique that measures regional cortical activity by measuring the potential difference at various points on the surface of the scalp. In this work the two brain imaging techniques are used to decode brain states, using a paradigm for three conditions: rest, motor and motor imagery.;The first part of the study attempts the classification of motor and motor imagery by using least square support vector machine (LS-SVM) with a radial basis function kernel. The data was recorded using functional near infrared spectroscopy. All pre- processing methods are selected to be possible for execution in a real-time setting. The first goal was to determine the optimal window length and starting point for the extraction of features. Once the optimal window length was established, two feature selection methods were compared: Fisher discriminant ratio (FDR) and the combined method, which uses FDR and K-means. Reducing the number of features improved the classification time with negligible impact on the classification accuracy.;The second part of the study uses a LS-SVM with a linear kernel to perform two classifications on EEG data: rest and motor imagery, and rest and motor. The average power of the frequency band between 10 Hz - 14 Hz was used to extract features from each channel. The two feature selection methods previously mentioned were compared. As expected the combined method produced better results.