Text

theses

Electricity is life but electricity is an invisible fist punching up your spine, knocking your brains right out of your skull." Ray Robinson One percent of the world's population suffers from epilepsy, a chronic disorder characterized by the occurrence of spontaneous seizures. About 30 percent of patients remain medically intractable and may undergo surgical intervention; despite the latter, some may still fail to attain a seizure-free outcome. The recent introduction of a closed loop system of localized electroencephalographic (EEG) recording and stimulus delivery (i.e., RNS, Neuropace) has provided greater opportunity to achieve control of this entity although further solutions are required to better actuate the system for optimal efficacy and to bring about an improved quality of life for these patients. A means of therapeutic interaction with an area of epileptogenicity, that does not entail removal of a portion of the brain, first requires adequate detection of ictal (seizure) onset. The use of computers to help physicians in the acquisition, management, storage, and reporting of brain (i.e., EEG) signals is well established. To this end, there are computer-aided detection applications that use a Brain Computer Interface (BCI). In order for a BCI system to work effectively, computational algorithms must reliably identify periods of increased probability of an impending ictal occurrence in order to abort its development. Such preictal periods may be of variable duration and may not afford suitable latency to provide current methodologies with sufficient time for signal deployment to achieve control in all circumstances. The development of an automated method that delivers on such short notice would optimize seizure control and bring about an improved quality of life. Technological innovation with BCI for control of epilepsy must acknowledge the immediacy of seizure occurrence and the time constraints imposed upon effective delivery. The goal of this PhD thesis is to design, develop and implement a BCI autonomic system for epileptic seizure detection and prediction using non-invasive and invasive brain big data. These include real-time data collection, data processing (e.g., feature extraction and classification) by a computer and biofeedback to apply the desired action. The requirements of a practical BCI system include methods for signal processing, machine learning, and brain-state prediction in large data sets collected from user populations in real-time and in combination with their health records. Learning applications of big data in the form of real-time acquisition with the background of the electronic healthcare record (EHR) provide for the generation of new knowledge that will aid in predictions of outcome and, therefore, prognosis. The current and future challenge for BCI centers upon developing methods and systems to remove noise, extract meaningful features and learn from big data. To address existing challenges, we introduce a new method for processing multimodal MRI, rs-fMRI, and EEG big data for epileptic ictal detection and prediction. Moreover, lateralization and epileptogenic network definition of the seizure focus are developed by multimodal processing of EEG/MRI big data. We develop a deep learning approach to extract high order features for seizure detection and prediction in epilepsy, leveraging the emerging mobile-edge computing platform. Finally, a systematic and statistical approach using a large dataset for the evaluation of automated segmentation methods in the epileptogenic hippocampus MRI segmentation is developed. In the first contribution of the thesis, a new method of feature selection and classification for rapid and precise seizure detection is developed wherein informative components of electroencephalogram~(EEG)-derived data are extracted and an automatic method is presented using Infinite Independent Component Analysis~(I-ICA) to select independent features. The feature space is divided into subspaces via random selection and multichannel Support Vector Machines~(SVMs) are used to classify these subspaces. The result of each classifier is then combined by majority voting to establish the final output. In addition, a random subspace ensemble using a combination of SVM, Multilayer Perceptron~(MLP) Neural Network and an extended k-Nearest Neighbors~(k-NN), called Extended Nearest Neighbor~(ENN), is developed for the EEG and Electrocorticography~(ECoG) big data problem. To evaluate the solution, a benchmark ECoG of eight patients with temporal and extratemporal epilepsy was implemented in a distributed computing framework as a multitier cloud-computing architecture. In the second contribution of the thesis, by leveraging the potential of cloud computing and deep learning, we develop and deploy BCI seizure prediction and localization from scalp EEG and ECoG big data. First, a new method for epileptic seizure prediction and localization of the seizure focus is presented. Second, an extended optimization approach on existing deep-learning structures, Stacked Auto-encoder and Convolutional Neural Network (CNN), is proposed based on principle component analysis (PCA), independent component analysis (ICA), and Differential Search Algorithm (DSA). Third, a cloud-computing solution (i.e., Internet of Things (IoT)), is developed to define the proposed structures for real-time processing, automatic computing and storage of big data. The proposed methods are compared for unsupervised feature extraction and big data classification of EEG channels to increase the accuracy of epileptic seizure prediction. A benchmark clinical dataset illustrates the superiority of the proposed patient-specific BCI as an alternative to current methodology to offer support for patients with intractable focal epilepsy. The key benefit of the proposed BCI centers upon the analysis and learning allowed from large amounts of unsupervised data, making it a practical method for developing a real-time patient-based seizure prediction and localization system. In the third contribution of the thesis, by leveraging the potential of autonomic edge computing in epilepsy, we develop and deploy both noninvasive and invasive methods for the monitoring, evaluation and regulation of the epileptic brain, with responsive neurostimulation (RNS; Neuropace). First, an autonomic edge computing framework is proposed for processing of big data as part of a decision support system for surgical candidacy. Second, an optimized model for estimation of the epileptogenic network using independently acquired EEG and rs-fMRI is presented. Third, an unsupervised feature extraction model is developed based on a convolutional deep learning structure for distinguishing interictal epileptic discharge (IED) periods from nonIED periods using electrographic signals from electrocorticography (ECoG). Experimental and simulation results from actual patient data validate the effectiveness of the proposed methods. In the fourth contribution of the thesis, a systematic and statistical approach using a large dataset for the evaluation of automated hippocampus segmentation methods is developed. Also, a method that would achieve results better approximating those attained by manual tracing in the epileptogenic hippocampus is established. A template database of 195 MR images of mTLE patients was used. Hippocampal segmentation was accomplished manually and by two well-known tools (FreeSurfer and Hammer) and two previously published methods developed at our institution (ABSS and LocalInfo). To establish which method was better performing for mTLE cases, several voxel-based, distance-based, and volume-based performance metrics were considered. Statistical validations of the results using automated techniques were compared with the results of benchmark manual segmentation. Extracted metrics were analyzed to find the method that provided a more similar result relative to the benchmark. The results confirmed that among the four automated methods, ABSS generated the most accurate results.

ETD_8858

Ph.D.

Includes bibliographical references

by Mohammad-Parsa Hosseini

doi:10.7282/T3VM4GQS

ETD doctoral

The author owns the copyright to this work.