A Modeling Approach to Analyze Epileptic Brain Data Using Univariate and Multivariate Dynamic Linear Models
MetadataShow full item record
Epileptic patients with severe cases of epilepsy usually undergo more than one surgery to locate and remove the epileptic tissue from the brain. It is hard to localize epileptic tissue as it is defined by abnormal excitability and synchronization of neurons, and does not have any distinguishing visual characteristics. During the first surgery a piece of cranium is removed and an array of EEG electrodes is placed on the surface of the brain, after which the patient is monitored for several days. During the second surgery, the brain is stimulated electrically and the result is monitored. Patients are traditionally monitored by EEG electrodes. However in the study done by Hochman and Haglund optical imaging technique is used to study the impact of the stimulus. In this project, we use data from a study conducted by Haglund and Hochman. The brain is divided into different regions of interest and illuminated with light of the desired wave-length, which is then photographed at intervals of approximately .2 to .3 seconds. The data is the series of average intensities for each region recorded at different times. Interpreting this time series data is challenging because of the noise contributed by heartbeat and respiration. We use a Bayesian dynamic linear model (DLM) to remove the noise artifacts. We first develop a univariate DLM to analyze each region independently. We observe that some regions responded to stimulus in a similar manner. To account for the correlation between these regions we further develop a multivariate DLM. These models help us to study the response of stimulus in noisy data, and will hopefully enable us to identify distinguishing properties of epileptic tissues in the brain.