Machine Learning for Analysis of Brain Signals / Event-Related Potential (ERP) measures derived from the electroencephalogram (EEG) have been widely used in outcome prediction of brain disorders. Recently, the ERPs that are transient (EEG) responses to auditory, visual, or tactile stimuli, have been introduced as useful predictors of a positive coma outcome (i.e. emergence from coma).
In this study, machine learning techniques were applied for detecting the Mismatch Negativity (MMN) component, which is a transient EEG response to auditory stimuli, and its existence has a high correlation with coma awakening, through analyzing ERPs signals recorded from healthy control brain signals. To this end, two different dimensionality reduction methods, Localized Feature Selection (LFS) and minimum-redundancy maximum-relevance (mRMR) were employed, where a localized classifier and the support vector machine (SVM) with radial basis function (RBF) kernel are used as classifiers. We trained both LFS and mRMR algorithms using signals of healthy brains and evaluated their performance for MMN detection on both healthy subjects and coma patients. The evaluation on healthy subjects, using leave-one-subject-out cross-validation technique, shows the detection accuracy performance of 86.6% (using LFS) and 86.5% (using mRMR).
In addition to analyzing brain signals for MMN detection, we also implemented a machine learning algorithm for discriminating healthy subjects from those who have experienced TBI. The EEG signals used in the TBI study were recorded using an ERP paradigm. However, we treated the recorded signals as resting state signals. To this end, we used the mRMR feature selection method and fed the selected features into the SVM classifier that outputs the estimated class labels. This method gives us a poor performance compared to the methods that directly used ERP components (without considering them as resting signals.). We conclude that our hypothesis of treating ERP data as resting data is not valid for TBI detection. / Thesis / Master of Applied Science (MASc)
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/25978 |
| Date | January 2020 |
| Creators | Arman Fard, Fatemeh |
| Contributors | Reilly, James, Connolly, John, Electrical and Computer Engineering |
| Source Sets | McMaster University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0023 seconds