Master of Science
De Ribaupierre, Sandrine
Epilepsy is a common neurological disorder that disrupts normal electrical activity in the brain causing severe impact on patients’ daily lives. Accurate seizure detection based on long-term time-series electroencephalogram (EEG) signals has gained vital importance for epileptic seizure diagnosis. However, visual analysis of these recordings is a time-consuming task for neurologists. Therefore, the purpose of this thesis is to propose an automatic hybrid model-based /data-driven algorithm that exploits inter-channel and temporal correlations. Hence, we use mutual information (MI) estimator to compute correlation between EEG channels as spatial features and employ a carefully designed 1D convolutional neural network (CNN) to extract additional information from raw EEGs. Then, seizure probabilities from combined features of MI estimator and CNN are applied to factor graphs to learn factor nodes. The performance of the algorithm is evaluated through measuring different parameters as well as comparing with previous studies. On CHB-MIT dataset, our generalized algorithm achieves state-of-the-art performance.
Summary for Lay Audience
Epilepsy is a common neurological disorder affecting about 50 million people worldwide. This disease is usually accompanied by the transient occurrence of signs or symptoms due to abnormal excessive electrical activity in the brain that may cause seizures. Our brain is constantly generating electrical pulses transmitted by neurons to control our movement, thoughts, and memories. In normal situations, neurons are firing independently or in small groups; however, during an epileptic seizure, many neurons fire simultaneously, 500 times faster than normal. The extensive sudden discharges in neural brain activity due to epileptic seizures scrambles the ’messages’ that the brain sends out to the rest of the body. This can lead to life-threatening consequences such as involuntary movements, sensations, and emotions and may cause a temporary loss of awareness and even death. As such, early detection of epileptic seizures can significantly improve quality of life for patients experiencing epileptic seizures. The most common technique used to diagnose seizures is reviewing scalp electroencephalogram (EEG) signals by a neurologist. Visually scanning the recordings, however, is a time-consuming process due to contamination by physiological and non-physiological resources as well as the similarity of seizure spikes to normal EEG wave-forms. As such, we have developed an automatic and generalized algorithm for seizure detection. To implement this approach, we capture two important underlying features that exist during seizure times, including computing inter-channel correlation among different EEG channels and temporal correlations between consecutive blocks. The first feature stems from the fact that when the seizure occurs in one or more parts of the brain, it propagates to other regions. This manifests itself as highly non-linear correlations among channels recording. The second property achieves due to spanning the seizure over multiple EEG blocks. Therefore, the thesis aims to propose a hybrid model-based/data-driven algorithm to exploit spatial and temporal correlations. The main two components in this algorithm are a neural mutual information estimator to compute inter-channel correlation and factor graph inference to exploit temporal correlations at reduced complexity. On the CHB-MIT dataset, our method obtains the best performance results compared to prior works.
Salafian, Bahareh, "Seizure Detection Using Deep Learning, Information Theoretic Measures and Factor Graphs" (2021). Electronic Thesis and Dissertation Repository. 8330.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.