Electroencephalography-based Imagery Movement Classification Using Support Vector Machine and Autoregressive Power Spectral Estimation

• Main Author: Pornwitcha Somsap
• Other Authors: Associate Professor Dr. Nipon Theera-Umpon
• Format: Theses and Dissertations
• Language: English
• Published: เชียงใหม่ : บัณฑิตวิทยาลัย มหาวิทยาลัยเชียงใหม่ 2020
• Online Access: http://cmuir.cmu.ac.th/jspui/handle/6653943832/69637
• Institution: Chiang Mai University

The main purpose of this thesis is about motor imagery classification to enhance communication capability for motor neuron disease patients, especially, those who cannot move their voluntary muscles such as Amyotrophic Lateral Sclerosis (ALS) or Locked-In Syndrome (LIS) patients. ALS is a progressive or damaged nervous system disease that affects nerve cells in the brain and spinal cord causing loss of muscle control. The symptoms of ALS are varied depending on the severity of damaged neurons. Signs and symptoms might include weakness, cramps, and twitching in the muscle. Likewise, the LIS patients cannot move their bodies except their eyes. Thus, ALS and LIS patients suffer to do necessary activities in daily living. According to the mentioned problems, this study focuses on the electroencephalography signal (EEG signal) because a patient’s brain still can function among the uncontrollable muscles. Additionally, the EEG signal can represent what the patient wants to communicate. Since an imagery hand movement is considered as a basic and simply thinking, it is used as the control activity for the experiment. The imagery left and right-hand movement can represent a basic answer (yes/no) or grasping (using the left or right hand). In this thesis, the EEG signal is collected and separated into 3 groups including imagery left-hand movement, right-hand movement, and relaxed state. There are 2 datasets, first is downloaded from a public EEG dataset and the other is recorded from a wireless EEG device. The signal is preprocessed and extracted features to build the model used for classification. The results show that the classification accuracy reaches almost 100% using the 5-fold cross validation.