Statistical Pattern Recognition Methods Applied to Evoked Electroencephalogram Data in Depressed Patients

Abstract

<p>Currently there exist a great deal of medication to deal with various mental disorders. Many of these medications serve similar purposes though only a select few work for a given individual. It can take several weeks to assess if any given medication is even working effectively. The purpose of this project is to develop an objective approach to diagnosing and more accurately treating various mental disorders with medications. To accomplish this one can observe relationships between surface currents of the brain and the patients mental disorder through statistical pattern recognition methods and appropriately assign a specific treatment using data a psychologist could not observe. The task has high computational requirements and made use of the McMaster electrical engineering grid. A framework to manipulate gigabytes of EEG data was established. A method to obtain features necessary was coded into the framework. Conducting a literary review of the field showed many similar depression calculations. Based on these depression feature calculations, an EEG analysis framework was established to easily obtain a number of studied features. Feature selection was used to find features which discriminate between classes. These features were then input into an support vector machine creating a classifier specifically designed for depression separation. A test data set of normals was used to perform relevant depression calculations including band powers, inter-hemispheric power ratios and coherences between all channel among other prominent calculations common throughout the EEG depression field. Results of the system were then analyzed to ensure accuracy and meaningfulness.</p>