EEG ABNORMALITY CLASSIFICATION BASED ON VISUALY ADOPTED QUANTITATIVE PARAMETER USING K-NEAREST NEIGHBOR AND DISCRIMINANT ANALYSIS

• Main Author: Sofiah - NIM: 23216131 , Amila
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/25452
• Institution: Institut Teknologi Bandung

Electroencephalogram (EEG) signal is electrical activity that occurs in the brain and is recorded by placing electrodes on the surface of the head. EEG reflects functional status of the brain. EEG signals possess unique and meaningful contribution to understanding the electrical functions of the brain. This signal is usually interpreted visually by electroencephalographers (EEGers) to diagnose diseases of the brain. Visual analysis in awake patients include the assessment of background rhythms and graphoelement. Normality and abnormality can be determined based on the assessment of these parameters. Unfortunately, visual analysis has several shortcomings. First, the lack of reliability because it depends on EEGer's capabilities and experience. Second, if spatial resolution of EEG increases, it requires longer time for assessment. Therefore, computer-based automated analysis can be an alternative to improve the reliability of assessment and to speed up the calculation process for large data. In this study an algorithm to quantify the EEG parameters was developed by adopting visual method used by clinicians in diagnosis. Since there was no clear threshold on quantitative parameters to classify normal and abnormal EEG signals, a probabilistic analysis was performed based on 14 features from the calculated parameters. Total of 130 normal and abnormal EEG data was used to train the algorithm. The results showed that the accuracy of EEG signal abnormalities classification by using 10fold cross validation with kNN and DA is 76.15% and 78.46%. In addition, the algorithm has a potential to classify specific cases. The kNN and LDA methods were able to classify epilepsy with a sensitivity larger than 50%. In conclusion, the extraction features of the visually adopted parameters were able to distinguish between normal and abnormal EEG signals. <br /> <br />