Date of Award
School of Engineering
Approximately one-third of patients diagnosed with focal epilepsy do not respond to medication and may be candidates for surgery to remove epileptogenic tissue known as the epileptogenic zone. A detailed pre-surgical evaluation is required and often includes invasive video electroencephalographic monitoring (IVEM) using intracranial surface and depth electrodes, and a camera. The resulting large pools of electrocorticorticographic (ECoG) data are manually analyzed by an expert epileptologist to determine epileptic events. The process is time consuming and prone to human error. This thesis investigates the use of measures to identify the causal relationship between ECoG signals during propagation of a seizure in order to delineate a possible epileptogenic zone. These measures are based on concepts of network connectivity derived from the frequency spectrum of recorded signals called the spectrumweighted directed transfer function (swDTF) and the full-frequency directed transfer function (ffDTF). The goal of the thesis is to implement a measure that may aid the surgeon in the decision-making process to optimize the outcome of surgery and possibly minimize the resection volume.
A time-variant adaptive version of both the swDTF and ffDTF was applied to a simple simulation model. The adaptive swDTF achieved higher sensitivity than the ffDTF (93% vs. 86%) for the detection of epileptogenicity. Both measures achieved a specificity of 99%. Two time-variant versions of the swDTFwere compared: 1) an adaptive approach to frequency spectrum estimation using a Kalman filtering algorithm and 2) a short-time spectral estimation approach using overlapping Hamming windows. Each method was successfully applied to a simple simulation model. The measures were then applied to electrodes of clinical ECoG data obtained from Spectrum Health’s Epilepsy Monitoring Unit. Sixteen seizures in two patients were analyzed and compared to channels indicated as having seizure activity by the epileptologist. The adaptive approach was able to identify the electrodes containing seizure activity consistent with expert findings (within 10 mm) in 14 out of 16 (88%) seizures. The short-time approach was able to identify an area within the region of interest (within 30-100mm) as noted by the epileptologist in 12 out of 16 (75%) seizures. The short-time swDTF reduced computation time by 95% compared to the adaptive approach. The short-time approach is more susceptible to noise and appears to be less selective whereas the adaptive approach is better able to pinpoint a single channel (± 10 mm). The adaptive measure is preferred due to its robustness to input parameters and ability to pinpoint channels. It is suggested that the short-time approach be used to gain quick insight into the region of interest identified by the 3-10 electrodes with the largest elevated output values and to later isolate single electrodes using the adaptive measure.
Gurisko, James Michael, "A Quantitative Tool for Identifying the Epileptogenic Zone using Network Connectivity Analysis" (2014). Masters Theses. 712.