Dissertations / Theses on the topic 'Epileptic seizures detection'

It is estimated that 65 million people worldwide have epilepsy, and many of them have uncontrollable seizures even with the use of medication. A seizure occurs when the normal electrical activity of the brain is interrupted by sudden and unusually intense bursts of electrical energy, and these bursts can be observed and detected by the use of an electroencephalograph (EEG) machine. This work presents an algorithm that monitors subtle changes in scalp EEG characteristics to predict seizures. The algorithm is built to calibrate itself to every specifc patient based on recorded data, and is computationally effcient enough for future on-line applications. The presented algorithm performs ICA-based artifact filtering and Lasso-based feature selection from a large array of statistical features. Classification is based on a neural network using Bayesian regularized backpropagation.The selected method was able to classify 4 second long preictal segments with an average sensitivity of 99.53% and an average specificity of 99.9% when tested on 15 different patients from the CHB-MIT database.

Epilepsy is one of the major neurological disorders that affects more than 50 million people around the world; it is characterized by unpredictable seizures due to an abnormal electrical activity in the brain. In this thesis nocturnal epilepsy has been investigated. In particular, Nocturnal Frontal Lobe Epilepsy (NFLE), that is a form of epilepsy in which seizures occur predominantly during sleep with symptoms including nocturnal awakenings, dystonic and tonic postures and clonic limb convulsions. The electroencephalographic (EEG) signals, which record the electrical activity of the brain, are used by neurologists to diagnose epilepsy. However, in almost 50% of NFLE cases, the EEG does not show abnormality during seizures, making the neurologists work to identify the epileptic events very difficult, thereby requiring the support of video recording to verify the epileptic events, with a subsequent time-consuming procedure. In literature few scientific contributions address the classification of nocturnal epileptic seizures. In this thesis, the automatic systems, both customized for single patient and generalized have been developed to find the best nocturnal epileptic seizure detection system from EEG signals. The combination of feature extraction and selection methods, associated to classification models based on Self Organizing Map (SOM), have been investigated following the classical machine learning approach. The ability of SOM to represent data from a high-dimensional space in a low-dimensional space, preserving the topological properties of the original space, has been exploited to identify nocturnal epileptic seizures and track the temporal projection of the EEG signals on the map. The proposed methods allow the definition of maps capable of presenting meaningful information on the actual brain state, revealing the mapping potential of clustering data coming from seizure and non-seizure states. The results obtained show that the patient-specific system achieves better performance than a patient-independent system. Moreover, comparing the performances with those of a binary classifier, widely used in epileptic seizure detection problems, the Support Vector Machine (SVM), the SOM model achieves good and, for some patients, higher performances. In particular, the patient-customized system using SOM model, reaches an average value of sensitivity and specificity equal to 82.85% and 89.92%, respectively; whereas the SVM classifier achieved an average sensitivity and specificity equal to 82.11% and 82.85%, respectively, suggesting the use of SOM model as a good alternative for nocturnal epileptic seizure detection. The discriminating power of SOM and the possibility to follow the temporal sequence of the EEG recordings on the map can provide information on an imminent epileptic seizure, highlighting the possibility to promote therapies aimed at rapid and targeted disarming the seizures.

In this work, we propose an approach that allows to explore the potential pathophysiological mechanisms (at neuronal population level) of ictogenesis by combining clinical intracranial electroencephalographic (iEEG) recordings with a neural mass model. IEEG recordings from temporal lobe epilepsy (TLE) patients around seizure onset were investigated. Physiologically meaningful parameters (average synaptic gains of the excitatory, slow and fast inhibitory population, Ae, B and G) were identified during interictal to ictal transition. We analyzed the temporal evolution of four ratios, i.e. Ae/G, Ae/B, Ae/(B + G), and B/G. The excitation/inhibition ratio increased around seizure onset and decreased before seizure offset, suggesting the disturbance and restoration of balance between excitation and inhibition around seizure onset and before seizure offset, respectively. Moreover, the slow inhibition may have an earlier effect on the breakdown of excitation/inhibition balance. Results confirm the decrease in excitation/inhibition ratio upon seizure termination in human temporal lobe epilepsy, as revealed by optogenetic approaches both in vivo in animal models and in vitro. We further explored the distribution of the average synaptic gains in parameter space and their temporal evolution, i.e. the path through the model parameter space, in TLE patients. Results showed that the synaptic gain values located roughly on a plane before seizure onset, dispersed during ictal and returned when the seizure terminated. Cluster analysis was performed on seizure paths and demonstrated consistency in synaptic gain evolution across different seizures from individual patients. Furthermore, two patient groups were identified, each one corresponding to a specific synaptic gain evolution in the parameter space during a seizure. Results were validated by a bootstrapping approach based on comparison with random paths. The differences in the path revealed variations in EEG dynamics for patients despite showing an identical seizure onset pattern. Our approach may have the potential to classify the epileptic patients into subgroups based on different mechanisms revealed by subtle changes in synaptic gains and further enable more robust decisions regarding treatment strategy. The increase of excitation/inhibition ratios, i.e. Ae/G, Ae/B and Ae/(B+G), around seizure onset makes them potential cues for seizure detection. We explored the feasibility of a model based seizure detection algorithm. A simple thresholding method was employed. We evaluated the algorithm against the manual scoring of a human expert on iEEG samples from patients suffering from different types of epilepsy. Results suggest that Ae/(B+G), i.e. excitation/(slow + fast inhibition) ratio, allowed the best performance and that the algorithm best suited TLE patients. Leave-one-out cross-validation showed that the algorithm achieved 94.74% sensitivity for TLE patients. The median false positive rate was 0.16 per hour, and median detection delay was -1.0 s. Of interest, the values of the threshold determined by leave-one-out cross-validation for TLE patients were quite constant, suggesting a general excitation/inhibition balance baseline in background iEEG among TLE patients. Such a model-based seizure detection approach is of clinical interest and could also achieve good performance for other types of epilepsy provided that more appropriate model, i.e. better describe epileptic EEG waveforms for other types of epilepsy, is implemented. Altogether, this thesis contributes to the field of epilepsy research from two perspectives. Scientifically, it gives new insights into the mechanisms underlying interictal to ictal transition, and facilitates better understanding of epileptic seizures. Clinically, it provides a tool for reviewing EEG data in a more efficient and objective manner and offers an opportunity for on-demand therapeutic devices.
Doctorat en Sciences de l'ingénieur et technologie
info:eu-repo/semantics/nonPublished

As a chronic neurological disorder, epilepsy is associated with recurrent, unprovoked epileptic seizures resulting from a sudden disturbance of brain function. Long-term monitoring of epileptic patients' Electroencephalogram (EEG) is often needed for diagnosis of seizures, which is tedious, expensive, and time-consuming. Also, clinical staff may not identify the seizure early enough to determine the semiology at the onset. This motivates EEG-based automated real-time detection of seizures. Apart from their possible severe side effects, common treatments for epilepsy (medication and surgery) fail to satisfactorily control seizures in ~25% of patients. EEG-based seizure prediction systems would significantly enhance the chance of controlling/aborting seizures and improve safety and quality of life for patients. This thesis proposes novel EEG-based patient-specific techniques for real-time detection and prediction of epileptic seizures and also presents a pilot study of scalp EEGs acquired in a unique low-noise underground environment. The proposed detection method is based on the wavelet packet analysis of EEG. A novel index, termed the combined seizure index, is introduced which is sensitive to both the rhythmicity and relative energy of the EEG in a given channel and considers the consistency among different channels at the same time. This index is monitored by a cumulative sum procedure in each channel. This channel-based information is then used to generate the final seizure alarm. In this thesis, a prediction method based on a variational Bayesian Gaussian mixture model of the EEG positive zero-crossing intervals is proposed. Novel indices of similarity and dissimilarity are introduced to compare current observations with the preictal and interictal references and monitor the changes for each channel. Information from individual channels is finally combined to trigger an alarm for upcoming seizures. These methods are evaluated using scalp EEG data. The prediction method is also tested against a random predictor. Finally, this thesis investigates the capability of an ultra-shielded underground capsule for acquiring clean EEG. Results demonstrate the potential of the capsule for novel EEG studies, including establishing novel low-noise EEG benchmarks which could be helpful in better understanding of the brain functions and mechanisms deriving various brain disorders, such as epilepsy.

L'objectif de ma thèse est d'analyser les caractéristiques des signaux inertiels et physiologiques qui générés par les mouvements inhabituels des patients lorsque la crise survient et de développer un algorithme pour détecter la crise. Notre approche dans le chapitre III commence par dériver la moyenne quadratique pour l'ACM et le Gyro, suivie de la normalisation de signaux entiers dans la même plage puis de l'agrégation en un seul signal. Le contrôle du graphique avec ses limites supérieure et inférieure est défini lors de la phase au repos et utilisé pour détecter les crises anormales et pour déclencher une alarme. La procédure dans le chapitre IV de détection s'exécute dans un dispositif de collecte de données portable et déclenche une alarme. Cet algorithme est basé sur la dérivation des mesures instantanées dans une plage de données glissante contenant des mesures inertielles de 3D (ACM), 3D Gyro et de EMG. La différence entre la puissance estimée et la puissance mesurée est utilisé comme entrée pour l'algorithme de détection basé sur la carte de contrôle de Shewhart. Lorsque la différence entre la puissance prévue et la puissance dérivée dépasse les limites [limite inférieure/supérieure] pour plusieurs créneaux consécutifs, une alarme est déclenchée. L'approche que nous proposons permet une bonne détection avec un FAR de 4\% et une sensibilité de 97\%. Notre modèle dans Le chapitre V commence par réduire la dimension des données collectées grâce à l'utilisation de la moyenne quadratique pour dériver un signal de 3D ACM et un signal du 3D Gyro. Avec les 3 signaux dérivés (ACM, Gyro et EMG), nous appliquons le TVP pour dériver un signal utilisé comme entrée pour le mécanisme de détection d'anomalie. La version robuste du z-score est appliquée sur le signal résultant produit pour détecter les déviations associées aux crises avant de déclencher une alarme. Nos résultats expérimentaux montrent que notre approche proposée est robuste contre les mouvements nocturnes et atteint un haut niveau de précision de détection avec un faible FAR. Ensuite, nous comparons les performances de notre approche avec la méthode des passages à zéro calculées à partir de sEMG. Notre approche montre que la précision de détection à l'aide du VTP surpasse le nombre de passages à zéro sur une plage glissante de chevauchement de 1 seconde. Dans le chapitre VI, Les appareils IoMT sont utilisés pour acquérir ACM, Gyro et EMG et pour transmettre les mesures à LPU pour traitement. Lorsque le LPU détecte des changements anormaux dans les mesures, il déclenche une alarme. Notre approche proposée utilise SVM avec option de rejet pour distinguer les crises des activités normales de la vie quotidienne. Les caractéristiques présentant des changements physiologiques de l'activité musculaire et les données inertielles ont été extraites dans LPU et sont utilisées comme entrée pour l'algorithme de détection. L'option de rejet dans SVM est utilisée pour améliorer la fiabilité du système de surveillance et pour réduire les fausses alarmes, où l'utilisateur est averti et a la possibilité de supprimer l'alarme dans son smartphone en l'absence de saisie. Les expériences menées ont prouvé que notre approche proposée peut atteindre une bonne précision pour distinguer les crises des activités normales avec seulement 4% de taux de FAR. Dans chapitre VII, nous proposons un cadre pour empêcher une MitM de perturber les opérations et interdire le déclenchement d'alarmes par le système de surveillance à distance des soins de santé. Pour réduire la consommation d'énergie lours de la transmission normale des données et préserver la confidentialité des données de santé, notre système transmet une signature de plus petite taille dérivée des données acquises avec un code d'authentification de message, où la clé est dérivée de RSSI. Nos résultats expérimentaux montrent que notre approche peut atteindre une précision de détection élevée avec un faible FAR de 3%
The goal of my PhD is to investigate the characteristics of inertial and physiological signals via IoMT systems generated by epileptic seizure and to develop an algorithm to detect the seizure. The focus of the algorithms lies in nocturnal seizures where the risk of SUDEP is high because the patients are unsupervised while sleeping. In chapter III analysis we propose an IoMT platform for seizure detection. The proposed framework approach starts by deriving the RMS for ACM and Gyro, followed by the normalization of whole signals (ACM, Gyro and EMG) in the same range, and aggregate all into one signal. The chart’s control with its upper and lower limits are derived in the training phase and used to detect abnormal seizures and to raise an alarm. In chapter IV Our proposed algorithm is based on deriving instantaneous power measurements in a sliding window containing 3D ACM or 3D Gyro or EMG. The residual between forecasted and measured power is used as input for the detection algorithm based on Shewhart Control Chart (SCC). When the difference between forecasted and derived power exceeds chart limits [lower, upper] for several consecutive slots, an alarm is raised. Our proposed approach provides low FAR (4%) and sensitivity of 97%. In Chapter V our proposed method starts by reducing the dimension of collected data using RMS to derive one signal from 3D ACM and one signal from 3D Gyro. With the derived 3 collected signals (ACM, Gyro and EMG), we apply VTP to derive one signal used as input for anomaly detection mechanism. The robust version of z-score is applied on the resulting product signal to detect deviations associated with seizures before raising an alarm. Our experimental results show that our proposed approach is robust against nocturnal movements and achieves a high level of detection accuracy with low false alarm rate. Afterward, we compare the performance of our approach with the zero-crossings method calculated from sEMG. Our approach shows that the detection accuracy using VTP outperforms zero-crossing count over an overlapping sliding window of 1 second. In chapter VI, we propose an approach using the IoMT devices to acquire EMG, ACM and Gyro data and to transmit the measurements to a LPU for processing. When the LPU detects abnormal changes in the measurements, it raises an alarm for assistant. Our proposed approach uses SVM with reject option to distinguish seizures from normal daily life activity. Features presenting physiological changes of muscular activity and inertial data were extracted in LPU and are used as input for the detection algorithm. The reject option in SVM is used to enhance the reliability of the monitoring system and to reduce FAR, where the user is notified and can discard the alarm in his smartphone in the absence of seizure. The conducted experiments proved that our proposed approach could achieve a good accuracy with only 4% of false alarm rate. Finally, since we are using IoMT sensors, which are susceptible to data security issues. We proposed a solution to prevent Man in the Middle (MitM) attack, which can identify healthcare emergencies of monitored patients and replay normal physiological data to prevent the system from raising an alarm. In this chapter, we propose a framework to prevent a MitM from disrupting the operations and prohibiting the remote healthcare monitoring system. To reduce energy consumption for normal data transmission, and preserve the privacy of health data, our framework transmits a smaller size signature derived from acquired data with message authentication code, where the key is derived from Received Signal Strength Indication (RSSI). Our experimental results for emergency detection show that our approach can achieve a high detection accuracy with a low false alarm rate of 3%

Epilepsy affects almost 1% of the global population and considerably impacts the quality of life of those patients diagnosed with the disease. Ambulatory EEG monitoring devices that can detect or predict seizures could play an important role for people with intractable epilepsy. Many outstanding studies in detecting and forecasting epileptic seizures using EEG have been developed over the past three decades. Despite this success, their implementations as part of implantable or wearable devices are still limited. To achieve high performance, many of these studies relied on handcraft feature extraction. This approach is not generalizable and requires significant modifications for each new patient. This issue greatly limits the applicability of such methods to hardware implementation. In this thesis, we propose a deep learning-based solution for generalized epileptic seizure detection and forecasting that does not require handcraft feature extraction. The method can be applied to any other patient without the need for manual feature extraction. Secondly, we optimize seizure detection and forecasting systems to reduce computational complexity and power consumption. The optimization is performed from two aspects: algorithm and input signal. In the first aspect, we propose two approaches: automatic channel selection to reduce the number of necessary EEG electrodes; Integer-Net, an integer convolutional neural network, to reduce computational complexity and required memory. In the second aspect, we investigate how sensitive seizure detection algorithms are regarding EEG's resolution. Another problem that we would like to address is the lack of labeled EEG data for epilepsy. Today the process of epileptic seizure identification and data labeling is done by neurologists, which is expensive and time-consuming. We propose an unsupervised learning approach to make use of unlabeled EEG data which is more accessible.

Epilepsy is one of the most common neurological diseases that affects millions of persons all over the world. The disease has always been of great importance in the biomedical field, due to the health risks it causes. It is characterized by recurrent, unprovoked seizures and can be assessed by the electroencephalogram (EEG). EEG measures the electrical activity in the brain, and one important aspect of the epilepsy research includes analyzing the EEG data in order to detect epileptic seizures in early stages. A lot of work has been done on patient-specific classifiers, but building patient-independent models is more difficult. This thesis focuses on the cross-patient view as it is more complicated due to EEG variability between different subjects. A comparative analysis of pattern recognition algorithms employed for EEG-based epileptic seizure identification was done. The algorithms compared was the Support Vector Machine (SVM) and K-Nearest Neighbor (KNN). Our study shows that the two methods perform similar, although KNN achieved a slightly higher accuracy during certain conditions.
Epilepsi är en av de vanligaste neurologiska sjukdomarna, vilken påverkar miljontals av människor över hela världen. Sjukdomen har alltid varit relevant inom det biomedicinska området på grund av hälsoriskerna den orsakar. Epilepsi karakteriseras av upprepade, oprovocerade anfall och kan fastställas med hjälp av elektroencefalografi (EEG). EEG mäter den elektriska aktiviteten i hjärnan, och en viktig aspekt inom epilepsiforskning inkluderar analys av EEG-data för att kunna detektera epileptiska anfall i ett tidigt skede. Mycket arbete har hittills gjorts på patient-specifika klassificeringsmetoder, medan det är svårare att bygga patient-oberoende modeller. Denna studie fokuserar på patient-oberoende klassificering eftersom den är mer komplicerad på grund av hur EEG-data skiljer sig mellan olika individer. En jämförelse av maskinlärningsmetoder för EEG-baserad detektion av epileptiska anfall utfördes. Algoritmerna som jämfördes var Support Vector Machine (SVM) och K-Nearest Neighbor (KNN). Vår studie visar att båda metoderna gav liknande resultat, dock uppnådde KNN en något högre noggranhet under vissa omständigheter.

Epilepsy has a significant adverse impact on almost 1% of people's health and well-being globally. Clinical EEG monitoring devices that enable seizure onset detection and prediction are crucial for epilepsy patients to manage their seizure disorders. In the past three decades, many epileptic seizure detecting, and prediction methods have been developed and reported high performance. However, most of them are retrospective and lack continental and multi-dataset generalization, transparency, and reproducibility, making them hard to implement into clinical utility. Besides, the seizure prediction biomarker is yet to be fully answered, and this issue significantly limits clinician trust when using the seizure prediction algorithms. In this thesis, we propose a generalized epileptic seizure detection AI-assisted system that tested on a large scale of the clinical EEG dataset and proved to improve time efficiency while accuracy alongside the human expert. The seizure detection performance is further improved by combining EEG and ECG using a novel multimodal AI system. Secondly, we propose a Bayesian convolutional neural network to facilitate the exploration of potential seizure forecasting biomarkers. Another problem we address is the need for long recording labeled EEG data for seizure prediction. We propose a novel real-time seizure prediction AI system that learns from the on-the-fly weak label generated by the detection model. Ultimately, we focus on developing a low-power, hardware-friendly implementation method using neuromorphic-compatible Spiking Neural Networks (SNNs) for seizure detection. Overall, the work presented in this thesis has tackled several research problems related to advanced AI applications in epileptic seizure detection and prediction and drove these emerging technologies toward building reliable AI systems in real-world clinical settings.

There have been numerous efforts in the field of electronics with the aim of merging the areas of healthcare and technology in the form of low power, more efficient hardware. However one area of development that can aid in the bridge of healthcare and emerging technology is in Information and Communication Technology (ICT). Here, databasing and analysis systems can help bridge the wealth of information available (blood tests, genetic information, neural data) into a common framework of analysis. Also, ICT systems can integrate real-time processing from emerging technological solutions, such as developed low-power electronics. This work is based on this idea, merging technological solutions in the form of ICT with the need in healthcare to identify normality in a patients’ health profile. In this work we develop this idea and explain the concept more thoroughly. We then go on to explore two applications under development. The first is a system designed around monitoring neural activity and identifying, through a processing algorithm, what is normal activity, such that we can identify anomalies, or abnormalities in the signal. We explore Epilespy with seizure detection and prediction as an application case study to show the potential of this method. The motivation being that current methods of prediction have proven to be unsuccessful. We show that using our algorithm we can achieve significant success in seizure prediction and detection, above and beyond current methods. The second application explores the link between genetic information and standard tests (blood, urine etc.) and how they link in together to define a personalised benchmark. We show how this could work and the steps that have been made towards developing such a database.

Epilepsy is the most common neurological disorder, affecting between 0.6% and 0.8% of the global population. Among those affected by epilepsy whose primary method of seizure management is Anti Epileptic Drug therapy (AED), 30% go on to develop resistance to drugs which ultimately leads to poor seizure management. Currently, alternative therapeutic methods with successful outcome and wide applicability to various types of epilepsy are limited. During an epileptic seizure, the onset of which tends to be sudden and without prior warning, sufferers are highly vulnerable to injury, and methods that might accurately predict seizure episodes in advance are clearly of value, particularly to those who are resistant to other forms of therapy. In this thesis, we draw from the body of work behind automatic seizure prediction obtained from digitised Electroencephalography (EEG) data and use a selection of machine learning and data mining algorithms and techniques in an attempt to explore potential directions of improvement for automatic prediction of epileptic seizures. We start by adopting a set of EEG features from previous work in the field (Costa et al. 2008) and exploring these via seizure classification and feature selection studies on a large dataset. Guided by the results of these feature selection studies, we then build on Costa et al's work by presenting an expanded feature-set for EEG studies in this area. Next, we study the predictability of epileptic seizures several minutes (up to 25 minutes) in advance of the physiological onset. Furthermore, we look at the role of the various feature compositions on predicting epileptic seizures well in advance of their occurring. We focus on how predictability varies as a function of how far in advance we are trying to predict the seizure episode and whether the predictive patterns are translated across the entire dataset. Finally, we study epileptic seizure detection from a multiple-patient perspective. This entails conducting a comprehensive analysis of machine learning models trained on multiple patients and then observing how generalisation is affected by the number of patients and the underlying learning algorithm. Moreover, we improve multiple-patient performance by applying two state of the art machine learning algorithms.

Thesis (Ph. D.)--Harvard-MIT Division of Health Sciences and Technology, 2009.
Cataloged from PDF version of thesis.
Includes bibliographical references (p. 157-162).
Epilepsy is a chronic disorder of the central nervous system that predisposes individuals to experiencing recurrent seizures. It affects 3 million Americans and 50 million people world-wide. A seizure is a transient aberration in the brain's electrical activity that produces disruptive physical symptoms such as a lapse in attention and memory, a sensory hallucination, or a whole-body convulsion. Approximately 1 out of every 3 individuals with epilepsy continues to experience frequent seizures despite treatment with multiple anti-epileptic drugs. These intractable seizures pose a serious risk of injury, limit the independence and mobility of an individual, and result in both social isolation and economic hardship. This thesis presents novel technology intended to ease the burden of intractable seizures. At its heart is a method for computerized detection of seizure onset. The method uses machine learning to construct patient-specific classifiers that are capable of rapid, sensitive, and specific detection of seizure onset. The algorithm detects the onset of a seizure through analysis of the brain's electrical activity alone or in concert with other physiologic signals. When trained on 2 or more seizures and tested on 844 hours of continuous scalp EEG from 23 pediatric epilepsy patients, our algorithm detected 96% of 163 test seizures with a median detection delay of 3 seconds and a median false detection rate of 2 false detections per 24 hour period.
(cont.) In this thesis we also discuss how our detector can be embedded within a low power, implantable medical device to enable the delivery of just-in-time therapy that has the potential to either eliminate or attenuate the clinical symptoms associated with seizures. Finally, we report on the in-hospital use of our detector to enable delay-sensitive therapeutic and diagnostic applications. We demonstrate the feasibility of using the algorithm to control the Vagus Nerve Stimulator (an implantable neuro stimulator for the treatment of intractable seizures), and to initiate ictal SPECT (a functional neuroimaging modality useful for localizing the cerebral site of origin of a seizure).
by Ali Hossam Shoeb.
Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2012.
Cataloged from PDF version of thesis.
Includes bibliographical references (p. 87-90).
This thesis addresses the problem of real-time epileptic seizure detection from intracranial EEG (IEEG). One difficulty in creating an approach that can be used for many patients is the heterogeneity of seizure IEEG patterns across different patients and even within a patient. In addition, simultaneously maximizing sensitivity and minimizing latency and false detection rates has been challenging as these are competing objectives. Automated machine learning systems provide a mechanism for dealing with these hurdles. Here we present and evaluate an algorithm for real-time seizure onset detection from IEEG using a machine-learning approach that permits a patient-specific solution. We extract temporal and spectral features across all intracranial EEG channels. A pattern recognition component is trained using these feature vectors and tested against unseen continuous data from the same patient. When tested on more than 875 hours of IEEG data from 10 patients, the algorithm detected 97% of 67 test seizures of several types with a median detection delay of 5 seconds and a median false alarm rate of 0.6 false alarms per 24-hour period. The sensitivity was 100% for 8 out of 10 patients. These results indicate that a sensitive, specific and relatively short-latency detection system based on machine learning can be employed for seizure detection tailored to individual patients. In addition, we describe and evaluate an algorithm for the detection of the cessation of seizure activity within IEEG. Seizure end detection algorithms can enable important clinical applications such as the delivery of therapy to ameliorate post-ictal symptoms, the detection of status epilepticus, and the estimation of seizure duration. Our machine-learning-based approach is patient-specific. The algorithm is designed to search for the termination of electrographic seizure activity once a seizure has been discovered by a seizure onset detector. When tested on 65 seizures, 88% of all seizure ends were detected within 15 seconds of the time determined by a clinical expert to represent the electrographic end of a seizure. We explore the effects of channel pre-selection on seizure onset detection. We evaluate and present the results from a seizure detector that has been restricted to use only a small subset of the channels available. These channels are manually chosen to be those that show the earliest ictal activity. The results indicate that performance can suffer in many cases when the algorithm uses a small set of selected channels, often in the form of an increase in false alarm rate. This suggests that the inclusion of a full channel set allows the system to leverage information that is not readily apparent to a clinical reader (from regions seemingly not involved in the onset) to better differentiate ictal and inter-ictal patterns. Finally, we present and evaluate an algorithm for patient-specific feature extraction, where the feature extraction process for a given patient leverages the training data available for that patient. The results from an evaluation of a detector that supplemented the original spectral energy features with features computed in a patient-specific manner show a significant improvement in 3 out of 5 patients. The results suggest that this is a promising avenue for further improvement in the performance of the seizure onset detector.
by Alaa Amin Kharbouch.
Ph.D.

Epilepsy is a chronic neurological brain disorder causing the affected to have seizures.Looking at EEG recordings, an expert is able to identify epileptic activity and diagnosepatients with epilepsy. This process is time consuming and calls for automatization. Theautomation process is done through feature extraction and classification. The featureextraction finds features of the signal and the classification uses the features to classify thesignal as epileptic or not. The accuracy of the classification varies depending on both whichfeatures is chosen to represent each signal and which classification method is used. Onepopular method for classification of data is the SVM. This report tests and analyses six featureextraction methods with a linear SVM to see which method resulted in best classificationperformance when classifying epileptic EEG data. The results showed that two differentmethods resulted in classification accuracies significantly higher than the rest. The waveletbased method for feature extraction got a classification accuracy of 98.83% and the Hjorthfeatures method got a classification accuracy of 97.42%. However the results of these twomethods was too similar to be considered significantly different and therefore no conclusioncould be drawn of which was the best.

Epilepsy affects up to 1% of the world's population and approximately 2.5 million people in the United States. A considerable portion (30%) of epilepsy patients are refractory to antiepileptic drugs (AEDs), and surgery can not be an effective candidate if the focus of the seizure is on the eloquent cortex. To overcome the problems with existing solutions, a notable portion of biomedical research is focused on developing an implantable or wearable system for automated seizure detection and control. Seizure detection algorithms based on signal rejection algorithms (SRA), deep neural networks (DNN), and neighborhood component analysis (NCA) have been proposed in the IoMT framework. The algorithms proposed in this work have been validated with both scalp and intracranial electroencephalography (EEG, icEEG), and demonstrate high classification accuracy, sensitivity, and specificity. The occurrence of seizure can be controlled by direct drug injection into the epileptogenic zone, which enhances the efficacy of the AEDs. Piezoelectric and electromagnetic micropumps have been explored for the use of a drug delivery unit, as they provide accurate drug flow and reduce power consumption. The reduction in power consumption as a result of minimal circuitry employed by the drug delivery system is making it suitable for practical biomedical applications. The IoMT inclusion enables remote health activity monitoring, remote data sharing, and access, which advances the current healthcare modality for epilepsy considerably.

Epilepsy is a chronic neurological disorder affecting around 50 million people worldwide. It is characterized by the occurrence of seizures; a transient clinical event caused by synchronous and/or abnormal and excessive neuronal activity in the brain. This thesis presents a novel machine learning toolbox that generates personalized epileptic seizure detection algorithms exploiting the information contained in electroencephalographic recordings. A large variety of features designed by the seizure detection/prediction community are implemented. This broad set of features is tailored to specific patients through the use of automated feature selection techniques. Subsequently, the resulting information is exploited by a complex machine learning classifier that is able to detect seizures in real-time. The algorithm generation procedure uses a default set of parameters, requiring no prior knowledge on the patients' conditions. Moreover, the amount of data required during the generation of an algorithm is small. The performance of the toolbox is evaluated using cross-validation, a sound methodology, on subjects present in three different publicly available datasets. We report state of the art results: detection rates ranging from 76% to 86% with median false positive rates under 2 per day. The toolbox, as well as a new dataset, are made publicly available in order to improve the knowledge on the disorder and reduce the overhead of creating derived algorithms.
L'épilepsie est un trouble neurologique cérébral chronique qui touche environ 50 millions de personnes dans le monde. Cette maladie est caractérisée par la présence de crises d'épilepsie; un événement clinique transitoire causé par une activité cérébrale synchronisée et/ou anormale et excessive. Cette thèse présente un nouvel outil, utilisant des techniques d'apprentissage automatique, capable de générer des algorithmes personnalisés pour la détection de crises épileptiques qui exploitent l'information contenue dans les enregistrements électroencéphalographiques. Une grande variété de caractéristiques conçues pour la recherche en détection/prédiction de crises ont été implémentées. Ce large éventail d'information est adapté à chaque patient grâce à l'utilisation de techniques de sélection de caractéristiques automatisées. Par la suite, l'information découlant de cette procédure est utilisée par un modèle de décision complexe, qui peut détecter les crises en temps réel. La performance des algorithmes est évaluée en utilisant une validation croisée sur des sujets présents dans trois ensembles de données accessibles au public. Nous observons des résultats dignes de l'état de l'art: des taux de détections allant de 76% à 86% avec des taux de faux positifs médians en deçà de 2 par jour. L'outil ainsi qu'un nouvel ensemble de données sont rendus publics afin d'améliorer les connaissances sur la maladie et réduire la surcharge de travail causée par la création d'algorithmes dérivés.

Submitted by Reginaldo Soares de Freitas (reginaldo.freitas@ufv.br) on 2017-08-22T13:26:59Z No. of bitstreams: 1 texto completo.pdf: 5760621 bytes, checksum: f90e38633fae140744262e882dc7ae5d (MD5)
Made available in DSpace on 2017-08-22T13:26:59Z (GMT). No. of bitstreams: 1 texto completo.pdf: 5760621 bytes, checksum: f90e38633fae140744262e882dc7ae5d (MD5) Previous issue date: 2017-03-10
Agencia Boliviana Espacial
A epilepsia é uma das doenças neurológicas mais comuns definida como a predisposição a sofrer convulsões não provocadas. A Organização Mundial da Saúde estima que 50 milhões de pessoas estão sofrendo esta condição no mundo inteiro. O diagnóstico de epilepsia implica em um processo caro e longo baseado na opinião de especialistas com base em eletroencefalogramas (EEGs) e gravações de vídeo. Neste trabalho, foram desenvolvidos dois métodos para a predição automática de convulsões usando EEG e mineração de dados. O primeiro sistema desenvolvido é um método específico para cada paciente (patient-specific) que consiste em extrair características espectro-temporais de todos os canais de EEG, aplicar um algoritmo de redução de dimensão, recuperar o envelope do sinal e criar um modelo usando um classificador random forest. Testando este sistema com um grande banco de dados de epilepsia, atingimos 97% de especificidade e 99% de sensibilidade. Assim, a primeira proposta mostrou ter um grande potencial para colaborar com o diagnóstico em um contexto clínico. O segundo sistema desenvolvido é um método não específico do paciente (non-patient specific) que consiste em selecionar o sinal diferencial de dois eletrodos, aplicar um vetor de bancos de filtros para esse sinal, extrair atributos de séries temporais e criar um modelo preditivo usando uma árvore de decisão CART. O desempenho deste método foi de 95% de especificidade e 87% de sensibilidade. Estes valores não são tão altos quanto os de métodos propostos anteriormente. No entanto, a abordagem que propomos apresenta uma viabilidade muito maior para implementação em dispositivos que possam ser efetivamente utilizados por pacientes em larga escala. Isto porque somente dois elétrodos são utilizados e o modelo de predição é computacionalmente leve. Note-se que, ainda assim, o modelo xigerado apresenta um poder preditivo satisfatório e generaliza melhor que em trabalhos anteriores já que pode ser treinado com dados de um conjunto de pacientes e utilizado em pacientes distintos (non-patient specific). Ambas as propostas apresentadas aqui, utilizando abordagens distintas, demonstram ser alternativas de predição de convulsões com performances bastante satisfatórias sob diferentes circunstâncias e requisitos.
Epilepsy is one of the most common neurological diseases and is defined as the pre- disposition to suffer unprovoked seizures. The World Health Organization estimates that 50 million people are suffering this condition worldwide. Epilepsy diagnosis im- plies an expensive and long process based on the opinion of specialist personnel about electroencephalograms (EEGs) and video recordings. We have developed two meth- ods for automatic seizure detection using EEG and data mining. The first system is a patient-specific method that consists of extracting spectro-temporal features of 23 EEG channels, applying a dimension reduction algorithm, recovering the envelope of the signal, and creating a model using a random forest classifier. Testing this system against a large dataset, we reached 97% of specificity and 99% of sensitivity. Thus, our first proposal showed to have a great potential for diagnosis support in clinical context. The other developed system is a non-patient specific method that consists of selecting the differential signal of two electrodes, applying an array of filter banks to that signal, extracting time series features, and creating a predictive model using a decision tree. The performance of this method was 95% of specificity, and 87% of sensitivity. Although the performance is lower than previous propos- als, due to the design conditions and characteristics, our method allows an easier implementation with low hardware requirements. Both proposals presented here, using distinct approaches, demonstrate to be seizure prediction alternatives with very satisfactory performances under different circumstances and requirements.

Tato práce se zabývá automatickou detekcí vysokofrekvenčních oscilací jakožto moderního elektrofyziologického biomarkru epileptogenní tkáně v intrakraniálním EEG, jehož vizuální detekce je zdlouhavý proces, který je ovlivněn subjektivitou hodnotitele. Epilepsie je jedním z nejčastějších neurologických onemocnění postihující 1 % obyvatelstva. Přestože jsou přibližně dvě třetiny případů léčitelné farmakologicky, zbylá třetina pacientů je odkázána zejména na léčbu chirurgickým zákrokem, pro nějž je zapotřebí přesně lokalizovat ložisko patologické tkáně. Vysokofrekvenční oscilace jsou v posledním desetiletí studovány pro jejich potenciál lokalizace patologické tkáně. Součástí této práce je shrnutí dosavadního výzkumu vysokofrekvenčních oscilací a výčet detektorů používaných ve výzkumu. V rámci práce byly vyvinuty či vylepšeny tři detektory vysokofrekvenčních oscilací, na jejichž popis navazuje evaluace z hlediska shody s manuální detekcí, přesnosti výpočtu příznaků oscilací a schopnosti lokalizace patologické tkáně. V závěru práce jsou představeny vyvinuté metody vizualizace vysokofrekvenčních výskytu oscilací a stručně uvedeny dosažené vědecké výsledky.

abstract: From time immemorial, epilepsy has persisted to be one of the greatest impediments to human life for those stricken by it. As the fourth most common neurological disorder, epilepsy causes paroxysmal electrical discharges in the brain that manifest as seizures. Seizures have the effect of debilitating patients on a physical and psychological level. Although not lethal by themselves, they can bring about total disruption in consciousness which can, in hazardous conditions, lead to fatality. Roughly 1\% of the world population suffer from epilepsy and another 30 to 50 new cases per 100,000 increase the number of affected annually. Controlling seizures in epileptic patients has therefore become a great medical and, in recent years, engineering challenge. In this study, the conditions of human seizures are recreated in an animal model of temporal lobe epilepsy. The rodents used in this study are chemically induced to become chronically epileptic. Their Electroencephalogram (EEG) data is then recorded and analyzed to detect and predict seizures; with the ultimate goal being the control and complete suppression of seizures. Two methods, the maximum Lyapunov exponent and the Generalized Partial Directed Coherence (GPDC), are applied on EEG data to extract meaningful information. Their effectiveness have been reported in the literature for the purpose of prediction of seizures and seizure focus localization. This study integrates these measures, through some modifications, to robustly detect seizures and separately find precursors to them and in consequence provide stimulation to the epileptic brain of rats in order to suppress seizures. Additionally open-loop stimulation with biphasic currents of various pairs of sites in differing lengths of time have helped us create control efficacy maps. While GPDC tells us about the possible location of the focus, control efficacy maps tells us how effective stimulating a certain pair of sites will be. The results from computations performed on the data are presented and the feasibility of the control problem is discussed. The results show a new reliable means of seizure detection even in the presence of artifacts in the data. The seizure precursors provide a means of prediction, in the order of tens of minutes, prior to seizures. Closed loop stimulation experiments based on these precursors and control efficacy maps on the epileptic animals show a maximum reduction of seizure frequency by 24.26\% in one animal and reduction of length of seizures by 51.77\% in another. Thus, through this study it was shown that the implementation of the methods can ameliorate seizures in an epileptic patient. It is expected that the new knowledge and experimental techniques will provide a guide for future research in an effort to ultimately eliminate seizures in epileptic patients.
Dissertation/Thesis
Doctoral Dissertation Electrical Engineering 2016

Dissertação de Mestrado em Engenharia Física apresentada à Faculdade de Ciências e Tecnologia da Universidade de Coimbra.
Epilepsy is a severe medical condition affecting millions of people in the world. Detection and prediction of epileptic events is an open problem and an active field of research in the medical and neuroscience communities. Patients with epilepsy are monitored at the service of Neurology from the Hospital de Santa Maria in Lisbon for long periods where huge amount of data, such as EEG, ECG and video, is collected in order to capture the appropriated number of epileptic episodes needed to characterize the disease. These episodes, impossible to predict, are random and sparse in time which makes the task of analysis difficult and time consuming. In this thesis an actigraphy device (an accelerometer) was designed and assembled to be plugged into the amplifier used in the hospital to collected the EEG signals. This way the activity of the patient on his non-dominant wrist (usually the left one) is recorded and stored simultaneously with the other signals acquired during the exam avoiding to perform the usual complex procedure of signal alignment and sampling frequency adjustment. Algorithms for detection of specific movement patterns were designed in order to automatically detect epileptic seizures with associated movement disturbances. The goal is to perform an automatic annotation of the data collected during exam and help the technical staff in its analysis. In the future these algorithms will be used in ambulatory systems to identify and record epileptic seizures in normal life conditions of the patients. Two prototypes were produced and tested at the hospital with real patients and the algorithms for movement detection and epileptic seizure identification were designed, implemented and tested using synthetic and real data. The EEG data from the patients were annotated and classified manually by the technicians where all the paroxysmal events were identified and used for training and testing as ground truth information. A total of 62 events from 4 patients, checked by the medical doctor, where used to train the detector and validate the algorithm. An accuracy of more than 98% was achieved in detecting movements and among them more than 84% were correctly classified as epileptic seizures. The two prototypes are installed at the hospital where more data are being collected.
A epilepsia é uma condição médica grave que afeta milhões de pessoas no mundo. A detecção e previsão de crises epilépticas é um problema em aberto e um campo ativo de pesquisa nas comunidades médica e de neurociência. Os pacientes com epilepsia são monitorizados no serviço de Neurologia do Hospital Santa Maria em Lisboa durante longos períodos, durante o qual uma grande quantidade de dados, tais como EEG, ECG e vídeo, é recolhida a fim de detectar o número necessário de crises epilépticas necessário para caracterizar a doença. Estes episódios, impossíveis de prever, são aleatórios e dispersos no tempo o que torna a tarefa de análise difícil e demorada. Nesta tese foi projetado e montado um dispositivo de actigrafia (um acelerómetro) para ser conectado ao amplificador usado no hospital para adquirir os sinais de EEG. Desta forma, a actividade motora do paciente é gravada e armazenada simultaneamente com os outros sinais adquiridos durante o exame evitando a realização do procedimento habitual e complexo de alinhamento de sinal e ajuste da frequência de amostragem, usando o dispositivo no pulso não-dominante (usualmente a esquerdo). Dois algoritmos para detecção de padrões de movimento específicos foram concebidos a fim de detectar automaticamente crises epilépticas com distúrbios de movimento associados. O objetivo é realizar uma anotação automática dos dados recolhidos durante o exame e ajudar os técnicos na análise. Futuramente, estes algoritmos iram ser utilizados em sistemas ambulatórios para identificar e registar as crises epilépticas em condições normais de vida dos pacientes. Dois protótipos foram produzidos e testados no hospital com pacientes reais e os algoritmos de detecção de movimento e identificação de crise epiléptica foram concebidos, implementados e testados utilizando dados sintéticos e reais. Os dados de EEG adquiridos nos pacientes foram anotados e classificados manualmente pelos técnicos nos quais foram identificados e utilizados para treino e testes todos os eventos paroxísticos. Um total de 62 eventos adquiridos a partir de 4 pacientes, verificados pelo médico, foram utilizados para treinar o detector e validar o algoritmo. Uma precisão de superior a 98% foi atingida na detecção de movimentos e entre eles mais de 84% foram corretamente classificados como crises epilépticas. Os dois protótipos estão instalados no hospital onde mais dados estão a ser adquiridos.

Αντικείμενο αυτής της εργασίας είναι η μελέτη τεχνικών για την ανάλυση δεδομένων που προέρχονται από συστήματα απεικόνισης της λειτουργίας του ανθρώπινου εγκεφάλου όπως το ηλεκτροεγκεφαλογράφημα. Σκοπός των τεχνικών ανάλυσης είναι η ανίχνευση συγκεκριμένων μορφών αυτών των σημάτων όπως για παράδειγμα οι επιληπτικές κρίσεις. Μία κρίση είναι μια παρέκκλιση στην ηλεκτρική δραστηριότητα του εγκεφάλου που παράγει αποδιοργανωτικά συμπτώματα για το άτομο και εκδηλώνεται κλινικά από εναλλαγή στη συμπεριφορά, στην κίνηση, στις αισθήσεις και στη συνειδητότητα. Οι κλινικές συμπεριφορές προηγούνται και στη συνέχεια συνοδεύονται από ηλεκτροεγκεφαλογραφικές αλλαγές. Η αυτόματη ανίχνευση των επιληπτικών κρίσεων μπορεί να αντιμετωπιστεί ως ένα πρόβλημα κατηγοριοποίησης των σημάτων σε κρίσεις ή όχι. Η ανίχνευση μπορεί να πραγματοποιηθεί σε δύο βήματα. Αρχικά εξάγονται χαρακτηριστικά που συλλαμβάνουν την μορφή και στη συνέχεια το διάνυσμα των χαρακτηριστικών δίνεται σε έναν εκπαιδευμένο κατηγοριοποιητή.
The subject of this work is the research of analysis techniques on data coming from neuroimaging systems such as Electroencephalogram. The aim of the data analysis techniques is the detection of specific morphologies of these signals such as the epileptic seizures. A seizure is a sudden breakdown of the neuronal activity of the brain that is clinically manifested by an involuntary alteration in behavior, movement, sensation, or consciousness. These clinical behaviors are preceded and then accompanied by electroencephalographic alterations. The automatic detection of epileptic seizures can be faced as a classification problem of the signals into seizures or non seizures. The detection can be carried out in two steps. Firstly, features which capture the morphology of the epileptic seizures are extracted and then the feature vector is given to an appropriately trained classifier.

Epilepsy is a serious neurological disorder characterized by recurrent unprovoked seizures due to abnormal or excessive neuronal activity in the brain. An estimated 50 million people around the world suffer from this condition, and it is classified as the second most serious neurological disease known to humanity, after stroke. With early and accurate detection of seizures, doctors can gain valuable time to administer medications and other such anti-seizure countermeasures to help reduce the damaging effects of this crippling disorder. The time-varying dynamics and high inter-individual variability make early prediction of a seizure state a challenging task. Many studies have shown that EEG signals do have valuable information that, if correctly analyzed, could help in the prediction of seizures in epileptic patients before their occurrence. Several mathematical transforms have been analyzed for its correlation with seizure onset prediction and a series of experiments were done to certify their strengths. New algorithms are presented to help clarify, monitor, and cross-validate the classification of EEG signals to predict the ictal (i.e. seizure) states, specifically the preictal, interictal, and postictal states in the brain. These new methods show promising results in detecting the presence of a preictal phase prior to the ictal state.

abstract: Approximately 1% of the world population suffers from epilepsy. Continuous long-term electroencephalographic (EEG) monitoring is the gold-standard for recording epileptic seizures and assisting in the diagnosis and treatment of patients with epilepsy. However, this process still requires that seizures are visually detected and marked by experienced and trained electroencephalographers. The motivation for the development of an automated seizure detection algorithm in this research was to assist physicians in such a laborious, time consuming and expensive task. Seizures in the EEG vary in duration (seconds to minutes), morphology and severity (clinical to subclinical, occurrence rate) within the same patient and across patients. The task of seizure detection is also made difficult due to the presence of movement and other recording artifacts. An early approach towards the development of automated seizure detection algorithms utilizing both EEG changes and clinical manifestations resulted to a sensitivity of 70-80% and 1 false detection per hour. Approaches based on artificial neural networks have improved the detection performance at the cost of algorithm's training. Measures of nonlinear dynamics, such as Lyapunov exponents, have been applied successfully to seizure prediction. Within the framework of this MS research, a seizure detection algorithm based on measures of linear and nonlinear dynamics, i.e., the adaptive short-term maximum Lyapunov exponent (ASTLmax) and the adaptive Teager energy (ATE) was developed and tested. The algorithm was tested on long-term (0.5-11.7 days) continuous EEG recordings from five patients (3 with intracranial and 2 with scalp EEG) and a total of 56 seizures, producing a mean sensitivity of 93% and mean specificity of 0.048 false positives per hour. The developed seizure detection algorithm is data-adaptive, training-free and patient-independent. It is expected that this algorithm will assist physicians in reducing the time spent on detecting seizures, lead to faster and more accurate diagnosis, better evaluation of treatment, and possibly to better treatments if it is incorporated on-line and real-time with advanced neuromodulation therapies for epilepsy.
Dissertation/Thesis
M.S. Electrical Engineering 2012

碩士
國立交通大學
電機資訊國際學程
107
Abstract Around 65 million people have been diagnosed with epilepsy. There are many ways offered for the treatment of epilepsy. The first treatment that doctor offer is Antiepileptic drugs. Antiepileptic drugs provide seizure control that is satisfying for most epilepsy patients but there are still 35 % of patient has uncontrolled epilepsy. Surgery could become the other option to remove the epileptogenic zone. However, the patients who did surgery still have to consume medicine and suffer seizures occasionally. These treatments still cause occasional seizures in patients, which affects the patient's quality of life and leads to accidents that may pose a danger to the patient and the people around them. From that intention to handle if the patient still has a seizure, this thesis develops epileptic seizure applications which process data from Epileptic Seizure Detection Tag (ESDT). These applications consist of two part: 1) ESDT APP; 2) Cloud service ESDT. The ESDT APP has features to send alert message and can record a daily record and epileptic seizure record which saves in the cloud. The alert system has a verification procedure to handle a false seizure before send SMS alert. As a complementary system, Cloud service ESDT has function to represent data from the cloud. Those applications record two kinds of data is daily EEG raw data and 30-second pre-ictal and ictal seizure signal which important for epileptic seizure analysis and improvement epileptic seizure detection algorithm. The verification data in applications is done by comparing pattern signal shape between data is received from ESDT which display on ESDT APP and data in the cloud which shows in the Cloud service ESDT. Data which is sent from on ESDT APP into the cloud can help patient record their seizure data and it is shown in Cloud service ESDT which help doctors for diagnosis or research. Keywords: Epileptic Seizure Application, Cloud, Web Application

碩士
國立交通大學
生醫工程研究所
107
Epileptic seizure control is always a popular issue. As time progresses, medication treatment and surgery are not the only way to control the symptom. Recently, there are there are much research about multi-channel seizure detection about multi-channel seizure detection. In order to get higher accuracy and stability, we require more number of channels. Increasing process for the channel may cause a heavy load of power consumption in the system. This thesis mentions that how to use the different channel characteristics to adjust the detecting time. The research extracts two kinds of features to be the based points of controlling detection. One is the frequency band power and the other one is the position of channel. The FFT calculates three bands of frequency power is the principal feature for predicting. Then each channel activity is classified by SVM model. Finally, the predictor decreases the detecting positions which includes numerous non-active channels. It protects from unwanted calculations which tends to decrease the power consumption efficiently. We build the new model for predicting. After simulation, the model can certainly decrease up to 45% of calculations and the seizure detecting still stay in high accuracy. In the future, the number of channels may continuously increase, and then the prediction system can bring more benefit.

碩士
國立交通大學
電子工程學系 電子研究所
104
Epileptic seizure control is a popular issue in recent years due to 30% of the epileptic patients remain drug-resistant and only some patients are suitable for resection surgery. The multi-channel seizure detection is also an important issue. The multi-channel detection can have more chances to cover the seizure onset zone so that the seizure can be suppressed efficiently. To achieve the better seizure control efficiency, the accuracy and the latency are necessary to reach certain levels (Accuracy > 90%, Latency < 5 s). In this thesis, a seizure detection algorithm with the training process and the simulation result is presented. The detection latency is 2.25s. For the data set, the sampling rate is 1024 Hz, 512 Hz or 256 Hz. However, in order to reduce the hardware complexity, the window length and the downsample issue are also simulated. The simulation result shows that with 1 s window and 128 Hz sampling rate, the accuracy can be up to 97.76%. A DSP processor for the 16-channel seizure detection has been designed and implemented. There are two main feature extraction circuits: 128-point approximate entropy and 128-point fast Fourier transform. The entropy block occupies 0.17mm2 while the FFT block occupies 0.58mm2, and the area of the DSP processor is 1.74mm2 in TSMC 0.18-um process. The operating frequency of the processor is 6.758 MHz and the power is 5.5 mW. To achieve better accuracy, the more complex algorithms are employed such as neural network (NN) and support vector machine (SVM). The simulation result shows that the multi-layer neural network can achieve the accuracy of 98.96% and the SVM is 99.25% so that the algorithms can provide a reliable detection results.

碩士
國立成功大學
電腦與通信工程研究所
98
In this thesis, we propose a new scheme which combines ICA with wavelet transformation on grouped EEG signals for epileptic seizure detection. The Independent Component Analysis (ICA) is adopted to enhance epileptic seizure. Then, wavelet transformation is followed with a dynamic threshold for identifying the epileptic seizure location. A series of experiments has been conducted to evaluate the proposed approach. The experimental results show that the proposed method has a superior performance than other approaches.

博士
國立臺灣大學
生醫電子與資訊學研究所
102
Epilepsy is a common chronic neurological disorder characterized by recurrent unprovoked seizures. The Electroencephalogram (EEG) signals play an important role in the diagnosis of epilepsy. In addition, multi-channel EEG signals have much more discrimination information than a single channel. However, traditional recognition algorithms of EEG signals are lack of multi-channel EEG signals. Due to large data computation, we propose a cloud based Epilepsy Analysis System (EAS) on multi-channel EEG signals. Both unipolar and bipolar EEG and ECG signals are both considered in our approach. We make use of approximate entropy (ApEn) and statistic values to extract features cascaded Genetic Algorithm (GA). Furthermore, EEG was also tested the performance by Support Vector Machine (SVM) and post-spike matching filters. We obtained accuracies of spikes and seizures are 86.69% and 99.77% for Clinical Data Set II. The detection system was further validated using the model trained by Clinical Data Set II on Clinical Data Set III. The system again showed high performance, with accuracies of spikes and seizures are 91.18% and 99.22%. Therefore, we built up a reliable, real-time, and complete (medical information and signal processing technology) system for detecting a large variety of seizures and spikes from multi-channel EEG data.

碩士
國立臺灣大學
生醫電子與資訊學研究所
100
Epilepsy is a common chronic neurological disorder characterized by recurrent unprovoked seizures. Seizure episodes can cause temporal paralysis of the body, which can lead to severe injuries. Electroencephalogram (EEG) is a tool commonly used for analyzing brain activity and diagnosing brain disorders. EEG can be presented under different montage schemes. This study focuses on two of the montage schemes; unipolar montage and bipolar montage. Traditionally, the most commonly used montage for automated EEG analysis is unipolar. We experiment with incorporating bipolar EEG montage for creating a classification system to classify different epileptic wave forms. A series of functions were designed for bipolar EEG montage. We used wavelet transform (WT) to decompose EEG signal into its primary sub-bands. We use Approximate Entropy and Total Variation as features designed specifically for spike and seizure detection. We used Genetic Algorithm and Fisher Score to rank and selected most influential features for classifier. Finally we use multi-class Support Vector Machine as our classifier.

碩士
國立交通大學
電機工程學系
106
Epilepsy is one of the most common neurological disorders; about 65 million people in the world are affected. A considerable portion of epilepsy cannot be well controlled by today’s available treatments. Even patients who have resection surgery to remove the epileptogenic zone will still suffer seizures once in a while. In this thesis, the design of a smart headband for epileptic seizure detection is presented. The proposed headband consists of four key components: 1) an analog front-end circuitry, 2) an epileptic seizure detection tag (ESDT), 3) a Bluetooth Low Power (BLE) chip, and 4) customized electrodes. All the above components are integrated into a fabric headband with only 50.3 g. The current consumption of the smart headband system is 16.35 mA. The epileptic seizure detection algorithm inside ESDT is validated by using Boston Children’s Hospital’s CHB-MIT scalp EEG clinical database with the detection rate of 92.68% and the false alarm of 0.527/hour. We develop a service APP connected to the cloud so that the patients’ health condition can be recorded and then referenced by doctors for further diagnosis or research. By event record, doctor only needs to review 1 % EEG recordings for precise diagnosis.

碩士
國立成功大學
資訊工程學系碩博士班
96
Epilepsy is one of the most common neurological disorders, approximately 1% of people in the world have epilepsy, 25% of epilepsy patients cannot be treated sufficiently by any available therapy. Epilepsy is caused by abnormal discharges in the brain, thus EEG has been an especially valuable clinical tool for the evaluation, detection, and treatment of epilepsy. Through EEG recordings, a number of systems which can release drug or give an electrical stimulation to suppress the seizures have been developed and under clinical operation for years. However, a robust device has not yet been developed which compute quickly and fast enough to action to meet immediately pathological changes of different types of seizures in human. In this paper, we propose a fast and reliable epilepsy detection method based on the complexity analysis and spectrum analysis. We propose complexity measure ApC and combine it with selected frequency bands power as the features for detecting seizures. An early seizure detection method is also presented which can detect seizures in a short time while seizures onset. Three different types of seizures are used for testing the detection performance. By the experiment result, the proposed epilepsy detection method can detect seizures in accuracy above 95% with a short detection delay 0.36-0.69 sec.

Tese de doutoramento em Ciência da Informação e Tecnologia, apresentada ao Departamento de Engenharia Informática da Faculdade de Ciências e Tecnologia da Universidade de Coimbra
This thesis concerns the problems of epileptic seizure prediction and detection. We analyzed multichannel intracranial electroencephalogram (iEEG) and surface electroencephalogram (sEEG) recordings of patients suffering from refractory epilepsy, to access the brain state in real time by using relevant EEG features and computational intelligence techniques, and aiming for detection of pre-seizure state (in the case of prediction) or seizure onset times (in the case of detection). Our main original contribution is the development of a novel relative bivariate spectral power feature to track gradual transient changes prior to ictal events for real-time seizure prediction. Furthermore a novel robust and generalized measure for early seizure detection is developed, aimed to be used in closed-loop neurostimulation systems. The development of a general platform embeddable on a transportable low-power-budget device is of utmost importance, for real time warning to patients and their relatives about the impending seizure or beginning of an occurring seizure. The portable device can also be integrated to work in conjunction with a closed-loop neurostimulation or fast-acting drug injection mechanism to eventually disarm the impending seizure or to suppress the just-occurring seizure. Therefore, in this thesis we try to meet the dual-objective of developing algorithms for seizure prediction and early seizure detection that provide high sensitivity and low number of false alarms, fulfilling the requirements of clinical applications, while being low computational cost. To seek the first objective, a patient-specific seizure prediction was developed based on the extraction of novel relative bivariate spectral power features, which were then preprocessed, dimensionally reduced, and classified using a machine-learning algorithm. The introduced feature bears low complexity, and was discriminated using the powerful support vector machine (SVM) classifier. We analyzed the preictal EEG dynamics across different brain regions and throughout several frequency bands, using relative bivariate features to uncover the underlying mechanisms ending in epileptic seizures. The suggested prediction system was evaluated on long-term continuous sEEG and iEEG recordings of 24 patients, and produced statistically significant results with average sensitivity of 75.8% and false prediction rate of 0.1 per hour. Furthermore a novel statistical method was developed for proper selection of preictal period, and also for the evaluation of predictive capability of features, as well as for the predictability of seizures. The method uses amplitude distribution histograms (ADHs) of the features extracted from the preictal and interictal iEEG and sEEG recordings, and then calculates a criterion of discriminability among two classes. The method was evaluated on spectral power features extracted from monopolar and bipolar iEEG and sEEG recordings of 18 patients, in overall consisting of 94 epileptic seizures. To approach the objective of early seizure detection, we have formulated power spectral density (PSD) of bipolar EEG signal in the form of a measure of neuronal potential similarity (NPS) between two EEG signals. This measure encompasses the phase and amplitude similarities of two EEG channels in a simultaneous fashion. The NPS measure was then studied in several narrow frequency bands to find out the most relevant sub-bands involved in seizure initiations, and the best performing ratio of two NPS measures for seizure onset detection was determined. Evaluating on long-term continuous iEEG recordings of 11 patients with refractory partial epilepsy (overall of 1785 h and 183 seizures) the results showed high performance, while requiring a very low computational cost. On average, we could achieve a sensitivity of 86.3%, a low false detection rate (FDR) of 0.048/h, and a mean detection latency of 14.2s from electrographic seizure onsets, while in average preceding clinical onsets by 1.1s. Apart from the above mentioned primary objectives, we introduced two new and robust methods for offline or real-time labelling of epileptic seizures in long-term continuous EEG recordings for further studies. Methods include mean phase coherence estimated from bandpass filtered iEEG signals in specific frequency bands, and singular value decomposition (SVD) of bipolar iEEG signals. Both methods were evaluated on the same dataset employed in the previous study and demonstrated sensitivity of 84.2% and FDR of 0.09/h for sub-band mean phase coherence, and sensitivity of 84.1% and FDR of 0.05/h for bipolar SVD, on average. Most of this work was established in collaboration with the EPILEPSIAE project, aimed to predict of pharmacoresistant epileptic seizures. The developed methods in this thesis were evaluated by the accessibility of long-term continuous multichannel EEG recordings of more than 275 patients with refractory epilepsy, referred to as The European Epilepsy Database. This database was collected by the three clinical centers involved in EPILEPSIAE, and contains well-documented metadata. The results of this thesis are backing the hypothesis of the predictability of most of epileptic seizures using linear bivariate spectral-temporal brain dynamics. Moreover, the promising results of early seizure detection sustain the feasibility of integrating the proposed method with closed-loop neurostimulation systems. We hope the developed methods could be a step forward towards the clinical applications of seizure prediction and onset detection algorithms.
Esta tese versa os problemas de predição e de deteção de crises epiléticas. Analisa-se o eletroencefalograma multicanal intracraniano (iEEG) e de superfície (sEEG) de pacientes que sofrem de epilepsia refratária, para a estimação em tempo real do estado cerebral, usando características relevantes do EEG e técnicas de inteligência computacional, ambicionando a deteção do estado pré-ictal (no caso de previsão) ou dos instantes de início de uma crise (no caso de deteção). A principal contribuição original é o desenvolvimento de uma característica de potência espectral bivariada relativa para captar as mudanças transitórias graduais que levam a crises e que poderão ser usadas para previsão em tempo real. Além disso, é desenvolvida uma nova medida, robusta e generalizada para a deteção precoce, destinada a ser utilizada em sistemas de neuro estimulação em malha fechada. O desenvolvimento de uma plataforma geral possível de ser integrada num dispositivo transportável, energeticamente económico, é de grande relevância para o aviso em tempo real do doente e dos seus próximos sobre a eminência da ocorrência de uma crise. O dispositivo transportável também pode ser usado em malha fechada com um neuro estimulador ou com um dispositivo de injeção rápida de um fármaco que desarme eventualmente a crise em curso. Por isso nesta tese persegue-se o objectivo de desenvolver algoritmos para previsão mas também para deteção de crises. Em ambos os casos, pretende-se que os algoritmos tenham uma elevada sensibilidade e uma baixa taxa de falsos positivos, tornando viável a sua utilização clínica. Para o objectivo de previsão, desenvolveu-se um método de previsão personalizado baseado na extração de uma característica nova, denominada de potência relativa espectral bivariada, que foi submetida a pre-processamento, redução de dimensão e classificação com Máquinas de Vetores de Suporte (SVM). Esta nova característica, de baixa complexidade, é computacionalmente simples, mas permite a análise da dinâmica do EEG preictal em diferentes regiões do cérebro e ao longo de várias bandas de frequência, de modo a descobrir os mecanismos subjacentes às crises epiléticas. O sistema de previsão obtido foi avaliado em registos contínuos de sEEG e iEEG de 24 pacientes, e produziu resultados estatisticamente significativos com sensibilidade média de 75.8% e taxa de predição falsa de 0.1 por hora. Além disso, foi desenvolvido um novo método estatístico para a seleção apropriada do período preictal, e também para a avaliação da capacidade preditiva das características, assim como para a própria previsibilidade das crises. O método utiliza os histogramas de distribuição de amplitude (ADHS) das características extraídas nos períodos pré-ictal e ictal dos registos de iEEG e sEEG e, em seguida, calcula um critério de discriminabilidade entre as duas classes. O método foi avaliado nas características de potencia espectral extraídas de registos iEEG e sEEG, monopolares e bipolares de 18 pacientes, consistindo num número total de crises epilépticas de 94. O segundo objetivo, a deteção precoce de crises, foi abordado através da formulação da densidade de potência espectral (PSD) de canais de EEG bipolares na forma de uma medida da similaridade do potencial neuronal (NPS) entre dois sinais de EEG. Esta medida usa as similaridades entre as fases e as amplitudes de dois canais de EEG de um modo simultâneo. A medida NPS foi estudada em várias bandas estreitas de frequência de modo a descobrir-se quais as sub-bandas mais envolvidas na inicialização das crises; buscou-se assim a melhor razão entre duas NPS do ponto de vista da deteção precoce. Avaliadas em iEEG contínuos de longa duração de 11 doentes com epilepsia refratária parcial (num total de 1785 h e 183 crises), os resultados apresentam um desempenho com sensibilidade de 86.3% e taxa de deteção falsa (FDR) de 0.048/h, uma latência de 14.2s em relação ao início eletrográfico, sendo uma crise detetada em média 1.1s antes da sua manifestação clínica. Para além dos objetivos principais referidos acima, introduziram-se dois novos métodos, robustos, para etiquetagem em diferido e em tempo real das crises em registos contínuos de EEG de longa duração para estudos posteriores. Esses métodos incluem a coerência de fase média (mean phase coherence) estimada a partir de registos iEEG em bandas de frequência específicas (usando filtros passa-banda), e a decomposição em valores singulares (SVD) de sinais iEEG bipolares. Ambos os métodos foram avaliados no mesmo conjunto de dados do estudo anterior e apresentaram, em média, uma sensibilidade de 84.2% e um FDR de 0.09/h para a coerência de fase média calculada para as sub-bandas, e sensibilidade de 84.1% e FDR de 0.05/h para a metodologia que usa a decomposição SVD bipolar. Grande parte deste trabalho foi feito no âmbito do projeto EPILEPSIAE, visando a previsão de crises em doentes epiléticos fármaco-resistentes. Os métodos desenvolvidos nesta tese aproveitaram a acessibilidade aos dados bem documentados de mais de 275 pacientes que constituem a Base de Dados Europeia de Epilepsia (European Epilepsy Database), provenientes dos três centros hospitalares participantes no projeto. Os resultados desta tese apoiam a hipótese da previsibilidade da maioria das crises epiléticas usando dinâmicas cerebrais bivariadas lineares espetrais e temporais. Além disso os resultados são promissores relativamente à deteção precoce de crises e sustentam a fazibilidade da integração desses métodos com técnicas de neuroestimulação em malha fechada. Esperamos que os métodos desenvolvidos resultem num avanço no que respeita à aplicação clínica de algoritmos de previsão e deteção de crises.
FCT - SFRH/BD/71497/2010

About 50 million people worldwide suffer from epilepsy and one third of them have seizures that are refractory to medication. In the past few decades, deep brain stimulation (DBS) has been explored by researchers and physicians as a promising way to control and treat epileptic seizures. To make the DBS therapy more efficient and effective, the feedback loop for titrating therapy is required. It means the implantable DBS devices should be smart enough to sense the brain signals and then adjust the stimulation parameters adaptively. This research proposes a signal-sensing channel configurable to various neural applications, which is a vital part for a future closed-loop epileptic seizure stimulation system. This doctoral study has two main contributions, 1) a micropower low-noise neural front-end circuit, and 2) a low-power configurable neural recording system for both neural action-potential (AP) and fast-ripple (FR) signals. The neural front end consists of a preamplifier followed by a bandpass filter (BPF). This design focuses on improving the noise-power efficiency of the preamplifier and the power/pole merit of the BPF at ultra-low power consumption. In measurement, the preamplifier exhibits 39.6-dB DC gain, 0.8 Hz to 5.2 kHz of bandwidth (BW), 5.86-μVrms input-referred noise in AP mode, while showing 39.4-dB DC gain, 0.36 Hz to 1.3 kHz of BW, 3.07-μVrms noise in FR mode. The preamplifier achieves noise efficiency factor (NEF) of 2.93 and 3.09 for AP and FR modes, respectively. The preamplifier power consumption is 2.4 μW from 2.8 V for both modes. The 6th-order follow-the-leader feedback elliptic BPF passes FR signals and provides -110 dB/decade attenuation to out-of-band interferers. It consumes 2.1 μW from 2.8 V (or 0.35 μW/pole) and is one of the most power-efficient high-order active filters reported to date. The complete front-end circuit achieves a mid-band gain of 38.5 dB, a BW from 250 to 486 Hz, and a total input-referred noise of 2.48 μVrms while consuming 4.5 μW from the 2.8 V power supply. The front-end NEF achieved is 7.6. The power efficiency of the complete front-end is 0.75 μW/pole. The chip is implemented in a standard 0.6-μm CMOS process with a die area of 0.45 mm^2. The neural recording system incorporates the front-end circuit and a sigma-delta analog-to-digital converter (ADC). The ADC has scalable BW and power consumption for digitizing both AP and FR signals captured by the front end. Various design techniques are applied to the improvement of power and area efficiency for the ADC. At 77-dB dynamic range (DR), the ADC has a peak SNR and SNDR of 75.9 dB and 67 dB, respectively, while consuming 2.75-mW power in AP mode. It achieves 78-dB DR, 76.2-dB peak SNR, 73.2-dB peak SNDR, and 588-μW power consumption in FR mode. Both analog and digital power supply voltages are 2.8 V. The chip is fabricated in a standard 0.6-μm CMOS process. The die size is 11.25 mm^2. The proposed circuits can be extended to a multi-channel system, with the ADC shared by all channels, as the sensing part of a future closed-loop DBS system for the treatment of intractable epilepsy.

碩士
國立成功大學
資訊工程學系碩博士班
97
Epilepsy is one of the most common neurological disorders, approximately 1% of people in the world have epilepsy, 25% of epilepsy patients cannot be treated sufficiently by any available therapy. Epilepsy is caused by abnormal discharges in the brain, thus EEG has been an especially valuable clinical tool for the evaluation, detection, and treatment of epilepsy. Through EEG recordings, a number of systems which can release drug or give an electrical stimulation to suppress the seizures have been developed and under clinical operation for years. However, a robust device has not yet been developed which compute quickly and fast enough to action to meet immediately pathological changes of different types of seizures in human. In this study, we research multi-type epilepsy diagnosis that can be applied to control the multi-type epilepsy with different method. We present a three-type epileptic diagnosis method, with using permutation entropy as the complexity index and spectrum band power with RBFSVM classify. The average accuracies of the RBFSVM reach to 79%. The detection rates of temporal EEGs can reach 97.6%. However, if we distinguish temporal EEGs (Set C) and non-temporal EEGs (Set A and B). The average accuracy can reach higher than 97% with RBFSVM. The classification results can be utilized to a system to determine what type of epilepsy when patients have mixed epilepsy. In the on-line detection, the results of the feasibility of developing algorithms to detect seizures based on automated analysis of the spatiotemporal dynamical characteristics of EEG recordings. We present a reliable epileptic seizures detection method, with using approximate entropy as the complexity index and spectrum band power. We present an adaptive threshold method that reduced false alarm. We also selecting recurrent neural network to be classifier which can detect seizures in a short time while seizures onset. The method has tested on three types of seizures including long-term recordings, robustly. This method was shown with several aspects of advantages, including high accuracy of on-line seizure detection (reach 100%), low false alarm (below 2.5%). The seizure detection latency was not greater than 0.5 sec after seizure onset.

碩士
國立交通大學
電機工程學系
102
Epilepsy is one of the most common neurological disorders, by which around 1% of the people in the world are affected. Unfortunately, 30% of the epilepsy patients cannot be treated sufficiently by antiepileptic drugs. As a result, a suitable implantable and portable medical device is a solution for epilepsy seizure control. To address these issues, an ASIC-based seizure detector and a RISC-based seizure detector are proposed to realize real-time closed-loop seizure detection algorithm. The ASIC-based seizure detector with wireless signal transmission is implemented in 0.18μm CMOS process and verified by animal experiment. The RISC-based seizure detector with FIR filter is implemented in 90 nm CMOS process and dissipates lower power compared with ASIC-based SoC when multi-channel seizure detection is realized.

碩士
國立交通大學
電信工程研究所
102
To improve the performance of epileptic seizure detection, independent component analysis (ICA) is applied to multi-channel signals to separate artifacts and signals of interest. FastICA is an efficient algorithm to compute ICA. To reduce the energy dissipation, eigenvalue decomposition (EVD) is utilized in the pre-processing stage to reduce the convergence time of iterative calculation of ICA components. EVD is computed efficiently through an array structure of processing elements running in parallel. Area-efficient EVD architecture is realized by leveraging the approximate Jacobi algorithm, leading to a 77.2% area reduction. By choosing proper memory element and reduced wordlength, the power and area of storage memory are reduced by 95.6% and 51.7%, respectively. The chip area is minimized through architectural transformations. Given the latency of 0.1s, an 86.5% area reduction is achieved compared to the direct-mapped architecture. Fabricated in 90nm CMOS, the core area of the chip is 0.40mm^2. The FastICA processor, part of an integrated epileptic control SoC, dissipates 81.6W at 0.32V. The computation delay of a frame of 256 samples for 8 channels is 84.2ms. Compared to prior work, 0.5% power dissipation, 26.7% silicon area, and 3.4 computation speedup are achieved. The performance of the chip was verified by human dataset.

Dissertação de Mestrado em Engenharia Informática apresentada à Faculdade de Ciências e Tecnologia
Conhecida como uma das doenças neurológicas mais comuns, a epilepsia afeta 1% da população mundial. Para 30% dos pacientes diagnosticados com epilepsia, não há tratamento ou medicação viável para evitar a ocorrência de crises. Esse tipo de epilepsia é chamado de epilepsia resistente a medicamentos. Isso significa que integridade física desses pacients pode estar comprometida a qualquer momento, o que pode colocar suas vidas em risco. Para essas pessoas, todo esforço bem-sucedido em prever ou detectar crises epilépticas tem o potencial de melhorar significativamente suas vidas.Esse assunto tem sido explorado na literatura bastante extensivamente, principalmente quando se trata de analisar características temporais e dados extraídos de vários electrodos. No entanto, existem muito poucos estudos que se concentram também em explorar as relações espaciais entre electrodos. A nossa ideia é explorar o potencial das relações espaciais entre os electrodos, através de mapas de electrodos, a fim de criar um modelo que possa prever ou detectar crises, possivelmente com melhor desempenho do que as já existentes. Nesse sentido, este projecto de tese aplicará técnicas de aprendizagem computacional profunda ao problema de detecção ou previsão de crises epilépticas, usando mapas de eletrodos derivados de dados do eletroencefalograma (EEG).Nosso modelo final é um detector de crises realista, que deve produzir alertas para crises em tempo real, usando redes neurais convolucionais em dados brutos de EEG. Esta tese explora os vários desafios da construção de um modelo capaz de alta sensibilidade e detecção precoce de um sistema como esse.Esta tese foi conduzida usando dados de EEG incluídos na European Epilepsy Database (banco de dados EPILEPSIA).
Known as one of the most common neurological disorders, epilepsy affects 1% of the world’s population. For 30% of epilepsy diagnosed patients, there is no viable treatment or medication to prevent the occurrence of seizures. This type of epilepsy is called drug resistant epilepsy. This means that their physical integrity is compromised, which eventually may put their lives at risk. For these people, every successful effort in predicting or detecting seizure events has the potential to significantly improve their lives.This subject has been explored in literature quite extensively, particularly when it comes to analyzing temporal features and data retrieved from various electrodes. However, there aver very few studies that focus also on exploring the spatial relations between electrodes. Our idea is to explore the potential of spatial relations between electrodes, through electrode maps, in order to create a model that can predict or detect seizures, possibly with better performance than already existing ones. Towards that end, this thesis project will be applying deep learning techniques to the problem of detecting or predicting epileptic seizures, using electrode maps derived from electroencephalogram (EEG) data.Our final model is a realistic seizure detector that is expected to produce alerts for seizures in real time using convolutional neural networks on raw EEG data. This thesis explores the various challenges of building a model capable of high sensitivity and early detection for a system like this.This thesis was conducted using scalp EEG data comprised in the European Epilepsy Database (EPILEPSIA database).

碩士
國立交通大學
電信工程研究所
100
Epilepsy is one of the most common neurological disorders, by which around 1% of the people in the world are affected. Unfortunately, 25% of the epilepsy patients cannot be treated sufficiently by antiepileptic drugs and epilepsy surgery. If seizures cannot be well controlled, the patients experience major limitations in their lives. In recent years, open-loop seizure controllers, such as vagus nerve and deep brain stimulation devices, have been proposed, but the effective rates of these devices are limited to 45%. In addition, low power and small hardware area are two important targets for implantable and portable devices. To overcome these issues, a real-time closed-loop seizure detection method is proposed. A multi-channel closed-loop epileptic seizure detector (MCESD) receives EEG signals of rats through ADC and delivers a stimulus at seizure. The seizure detection algorithm is realized by MCESD. The MCESD is implemented in a TSMC 0.18μm CMOS process. The seizure detection accuracy of device is above 94.6% from seizure detection algorithm with MCESD implementation, and the power of chip consumes 114.4μW.

博士
國立交通大學
電信工程研究所
101
In recent years, integrated circuits play an important role in today’s personal medical applications. These medical devices are usually battery-powered and their limited power budget imposes design challenges on signal acquisition and processing. Epileptic seizure control is one of the emerging applications. In this application, closed-loop neurostimulation is the most important method for seizure control devices. A real-time seizure detector is the kernel of a closed-loop seizure controller. In this dissertation, several low-power high-performance techniques from software to hardware level are applied for real-time power-efficient seizure detection. To demonstrate the proposed ideas, three works are designed and implemented. Long-Evans rats with spontaneous absence seizures are used as animal models for long-term continuous verification. In the first work, a bio-signal processor (BSP) core based on 32-bit reduced instruction set computer (RISC) architecture for seizure detection is implemented to achieve low-power consumption and continuous real-time processing. The proposed BSP core consists of 5-stage integer pipeline, 32×32 multiply-accumulator (MAC) unit, and a 32-bit tick timer. These features can enable high-performance signal processing and task scheduling for many biomedical applications. The floating-point seizure detection algorithm is approximated and rescheduled for short latency. The high-performance BSP core is implemented in 0.18 ?慆 complementary-metal-oxide semiconductor (CMOS) technology to verify functionality and capability. The measurement results show that the implemented processor can reduce over 90% power consumption compared with our previous prototype, which is implemented on an enhanced 8051 microcontroller. In the second work, a power-efficient BSP based on the first work is proposed to utilize for diverse physiological signals. Tens of kilobytes memory is embedded for efficient program execution in the proposed processor. The multi-mode analog-to-digital converter (ADC) is also integrated for physiological signals acquisition. Several serial and parallel ports are integrated with RISC processor for system expansion. Significant performance improvement is achieved through instruction optimization. Voltage and frequency scaling as well as clock gating are applied to reduce dynamic power on this work. The proposed BSP is implemented in 0.18 ?慆 CMOS technology. The measurement results show that the BSP consumes hundreds of microwatts to perform real-time seizure detection. The highly integrated and power-efficient BSP can be applied for excessive portable medical devices. The last work presents a power-efficient seizure detection system-on-chip (SoC). The FFT and entropy coding engines with direct memory access (DMA) feature are designed to reduce dynamic power through high-performance computation. The sample buffer and data control unit for signal acquisition is proposed to reduce context switching overhead. The seizure detection SoC is implemented in 0.18 ?慆 CMOS technology. The simulation results show that the implemented SoC consumes tens of microwatts to perform real-time seizure detection. The ultra-low power consumption of the proposed SoC enables implantable closed-loop seizure suppression in the future. Combining with efficient hardware architecture and software optimization, the real-time processing capability, design flexibility, portability, and versatility of the proposed platform and its design methodology can be applied on closed-loop seizure controller and many biomedical implants.

碩士
國立交通大學
電信工程研究所
100
Epilepsy is one of the most common neurological disorders. Approximately 1% of people in the world suffer from epilepsy, and 25% of epilepsy patients cannot be healed by today’s available treatments. In past years, open-loop seizure controllers have been proposed, such us vagus nerve stimulation and deep brain stimulation devices; however, the device drives a stimulator continuously or intermittently that causes high power consumption and the likelihood of neuronal damage. In contrast, the closed-loop implementation of hardware prototypes or biomedical signal processors has been proposed recently. Nevertheless, the average of seizure detection delay is either longer than 5 seconds or often not mentioned in these works, and it is insufficient to validate the robustness of detection algorithm. Moreover, most of studies often use the discontinuous electroencephalogram (EEG) signal fragments to validate seizure detection algorithm. As a result, a portable wireless online closed-loop seizure controller in freely moving rats was proposed, which validated seizure detection algorithm by using continuous online EEG signals. In this thesis, the fast parameter determination method, which determines a fitting model for each rat, is proposed to improve our previous work. The proposed parameter determination method is 416*10E6 times faster than our previous work, and it can attain the same detection accuracy (92-99%) and detection delay (0.63-0.79 s). Additionally, a low-power biomedical signal processor which bases on reduced instruction set computer (RISC) technology consumes only 6 mW for real-time epileptic seizure detection algorithm. Compared with our previous prototype, the measurement results show that the implemented processor can reduce 93.8% power consumption. The developed seizure detector can be applied to monitor the online EEG signals and integrate with analog front-end circuitries and an electrical stimulator to perform a closed-loop seizure controller in the future.

Trabalho de Projeto do Mestrado Integrado em Engenharia Biomédica apresentado à Faculdade de Ciências e Tecnologia
Epilepsy is a non-communicable brain disease that affects approximately 50 million people of all ages worldwide. It is estimated that about 2.4 million people are diagnosed with epilepsy per year. As a way to improve their lives, it is important to understand the changes occurred during and especially before crises to enable their correct prediction.The purpose of this study was the development of a computational application able to analyze and process electrochemical signals recorded in vivo using platinum microelectrodes arrays microfabricated in ceramic support chronically implanted into the brain of rats treated with pilocarpine as temporal lobe epilepsy model and displaying suitable electroanalytical properties for the measurement of oxygen with high resolution in the brain extracellular space.In this study, extraction of univariated features of the collected signal is executed based on the analysis of electroencephalograms (EEG). The developed application implements a processing chain composed of: pre-processing; extraction of characteristics; reduction / selection of characteristics; classification and analysis of results.Four Wistar rats were studied at eight weeks of age, showing that it is possible to identify the moments in which seizures occur through analyzed features. The mean sensitivity and specificity in detection test data were 66.41% and 50.4%, respectively. However, for some animals a sensitivity and specificity above 80% have been detected.Through the application, it was possible to observe that different pre-processing options introduce changes on data classification performance, although a concrete pattern is not visible, probably due the small number of analyzed animals. To the best of our knowledge it appears to be no seizure prediction/detection studies using pO2 signals, which makes the results presented in this thesis innovative.
A epilepsia é uma doença cerebral crónica não transmissível, que afeta aproximadamente 50 milhões de pessoas de todas as idades em todo o mundo. Estima-se que, por ano, são diagnosticadas cerca de 2,4 milhões de pessoas com epilepsia. Como forma de melhorar a vida destas pessoas é importante perceber as alterações ocorridas durante e principalmente antes das crises, para possibilitar a sua previsão no futuro e consequentemente a melhoria da qualidade de vida.Esta tese descreve o desenvolvimento de uma aplicação computacional com a capacidade de analisar e processar sinais eletroquímicos recolhidos através de microeléctrodos de platina em matriz microfabricados num suporte cerâmico e que possuem as propriedades eletroanalíticas adequadas para a medição de oxigénio in vivo, no espaço extracelular no cérebro, com elevada resolução espacial e temporal, depois de implantados no cérebro de ratos tratados com pilocarpina como modelo de epilepsia do lobo temporal.No estudo é realizada a extração de características univariadas dos sinais recolhidos com base em estudos prévios realizados com eletroencefalogramas (EEG). A aplicação desenvolvida implementa uma cadeia de processamento composta por: pré-processamento; extração de características; redução/seleção de características; classificação e análise de resultados.Foram estudados quatro ratos Wistar com 8 semanas de vida, mostrando que é possível identificar os momentos em que ocorrem crises através das características analisadas. A sensibilidade e especificidade médias em dados de teste de deteção foram de 66,41% e 50,04%, respetivamente. No entanto, para alguns animais obtiveram-se sensibilidades e especificidades acima de 80%.Foi possível perceber que diferentes opções de processamento introduzem alterações no desempenho da classificação dos dados, apesar de não ser visível um padrão concreto, provavelmente devido ao reduzido número de animais analisados.No melhor do nosso conhecimento parece não haver estudos de previsão/deteção de crises usando sinais relativos à pO2, o que torna os resultados apresentados nesta tese inovadores.