Rozprawy doktorskie na temat „EEG SINGLE-CHANNEL”

The implementation of portable electroencephalography (EEG) systems has been known to be complex. During ambulation, the integrity of recorded EEG signals is often impaired by motion artifacts and the time and effort required to set up the system is excessive. The use of single-channel EEG systems with dry, active electrodes (AEs) for signal acquisition is a topic of current interest. AEs are electrodes which have integrated bioamplifier circuitry and are known to be less susceptible to motion artifacts and environmental interference. In this report, the design of an AE preamplifier for the purpose of improving single-channel EEG recordings is presented. Initially, a thorough literature review was performed, exploring the available knowledge and state-of-the-art technology. Thereafter, the design specifications were set and the appropriate topology and circuit design techniques were selected to maximize the amplifier’s performance. Ultimately, three different preamplifier topologies were designed and their performance compared with one another as well as with established medical device standards and state-of-the-art AEs. The results of one preamplifier showed comparable performance with state-of-the-art AEs. Therefore, this topology was selected for a deep analysis and physical layout design. The layout of the selected preamplifier was designed and its parasitics extracted. The post layout performance of the design proved to be comparable to the schematic level performance, with a CMRR of 153dB, IRNV of 0.89µVRMS and an electrode offset tolerance of 450mV. The preamplifier design presented in this report has proven to be comparable with state-of-the-art AE preamplifiers and demonstrates potential for the advancement of AE performance in single-channel EEG systems.
Implementeringen av bärbara elektroencefalografisystem (EEG) har varit känd för att vara komplex. Vid rörelse påverkas ofta reliabiliteten av de inspelade EEGsignalerna av rörelseartefakter samt av att tiden och det arbete som krävs för att ställa in systemet blir överdrivet lång. Användandet av singelkanals EEG-system, med torra aktiva elektroder (AE) under inspelningen, är ett aktuellt ämne. En AE är en elektrod som har en integrerad bioförstärkarkrets och är känd för att vara både mindre mottaglig för rörelseartefakter och för störning från omgivningen. I denna rapport presenteras utformningen av en AE-förförstärkare för singelkanals EEGinspelningar. Inledningsvis utfördes en grundlig litteraturöversikt där den rådande kunskapen och toppmoderna tekniken undersöktes. Därefter bestämdes designspecifikationerna, lämpliga topologier samt kretsdesigntekniker, för att maximera förstärkarens prestanda. Slutligen konstruerades tre olika förstärkares topologier och deras prestanda jämfördes med varandra liksom med etablerade medicin tekniska standarder och toppmoderna AE. Resultaten av en förförstärkare visade sig ha jämförbar prestanda med toppmoderna AE. Därför valdes denna topologi ut för en djupanalys och för fysisk layoutkonstruktion. Layouten för den valda förförstärkaren utformades och dess parasiter extraherades. Utformningen av postlayouten visade sig vara jämförbar med prestanda på en schematisk nivå, med en CMRR på 153dB, IRNV på 0.89µVRMS och en elektrodoffsettolerans på 450mV. Förförstärkarens design som presenteras i denna rapport har visat sig vara jämförbar med toppmoderna AE-förförstärkare och visar potential till framsteg för AEprestanda i singelkanals EEG-system

Driver fatigue is a state of reduced mental alertness which impairs the performance of a range of cognitive and psychomotor tasks, including driving. According to the National Highway Traffic Safety Administration, driver fatigue was responsible for 72,000 accidents that lead to more than 800 deaths in 2015. A reliable method of driver fatigue detection is needed to prevent such accidents. There has been a great deal of research into studying driver fatigue via electroencephalography (EEG) to analyze brain wave data. These research works have produced three competing EEG data-based ratios that have the potential to detect driver fatigue. Research has shown these three ratios trend downward as fatigue increases. However, no empirical research has been conducted to determine whether drivers begin to feel fatigue at a certain Percent Change from an alert state to a fatigue state in one or more of these ratios. If a Percent Change could be identified for which drivers begin to feel fatigue, then it could be used as a method of fatigue detection in real-time system. This research focuses on answering this question by collecting brain wave data via an EEG device over a 60-minute driving session for 10 University of North Florida (UNF) students. A frequency distribution and cluster analysis was done to identify a common Percent Change for the participants who experienced fatigue. The results of the analysis were compared to a subset of users who did not experience fatigue to validate the findings. The project was approved by the UNF IRB on Nov. 1, 2016 (reference number 475514-4).

Poor sleep measured across many dimensions has been linked to adverse physical and mental health outcomes including cardiovascular disease, diabetes, cancer, increased mortality, depression, and anxiety. Current research typically relies upon brief, subjective, inadequately validated methods to assess limited dimensions of sleep, resulting in inaccurate measurements and possibly faulty conclusions. Specifically, research validating objective (e.g., actigraphy) and subjective (e.g., sleep diaries, retrospective surveys) measurement methods against the gold standard of polysomnography (PSG, an overnight sleep study) is primarily limited by a) a lack of reliability based on too short (e.g., 24 or 48 hours) of an assessment period to capture night-to-night variability, b) a lack of ecological validity (e.g., full PSG in a laboratory setting), and c) a lack of generalizability due to limited or special populations (e.g., individuals with insomnia). Barriers such as prohibitive cost, extensive setup time, and personnel training requirements diminish the ability of researchers to conduct measurement comparison studies using gold standard measures like traditional PSG. These barriers can be circumvented with the use of low-cost, minimally invasive single-channel EEG devices (e.g., Zmachine), but to date few studies have employed these devices. The current study evaluated the accuracy of retrospective surveys, sleep diaries, and actigraphy compared to a single-channel EEG device for assessment of sleep timing, duration, and efficiency in participants' homes over one week using a broad community sample (N = 80). Actigraphy generally demonstrated the best agreement with Zmachine across sleep variables, followed by diary and then survey. Circadian midpoint was the most consistent across measures, followed by sleep duration and then sleep efficiency. Implications and future directions are discussed.

碩士
國立陽明大學
醫學工程研究所
97
The quality of sleep is effective to our daily life, and more and more people have been experiencing sleep disorders in recent years. However the equipment and resources for sleep diagnosis are still limited. Although many researchers have developed automatic sleep scoring system, most of the systems are built based on PSG recording. A new homecare sleep recording device, “VitalBelt”, which records only three channel bio-signals (EEG, EOG, ECG), is applied in our scoring system. Sleep features are extracted individually, then a scoring method including exact scoring, refinement and impossible rejection three aspect to score sleep stages. Thirty subjects were recorded using both polysnomnography and VitalBelt. Comparing the manual score by polysnomography and automatic score by VitalBelt, overall agreement was 73.48% (Cohen’s kappa = 0.62). The result showed that the automatic sleep staging system developed for VitalBelt was reliable and might apply in homecare sleep monitoring. In addition, the methods of single-channel EEG with or without EOG were compared to evaluate the requirements of EOG for sleep staging. Paired t-test was used to estimate the agreement increase, with the result showing that the improvement was significant (P < 0.005) after combining EOG.

碩士
義守大學
電子工程學系
106
An emotional stress determination system is helpful for the modern people to reduce stress and avoid conflict, anxiety and misunderstanding. In this study, we used different multimedia to induce emotions and obtain EEG data. The raw EEG data was transformed into different EEG types by Fast Fourier Transform (FFT). However, the inducing emotion effect from subjects might be different because of the personal preferences, values and life experiences. Therefore Self-Assessment Manikin (SAM) is used as criterions for examining subjects’ emotion after induced emotion experiment. Experimental results shows that Support Vector Machine (SVM) achieves emotional determination with an average accuracy of 86.4%.The finding of this work is the high-gamma(γ) frequency feature from EEG power has the higher average accuracy rate of both positive emotion and negative emotion.

碩士
元智大學
機械工程學系
91
The control system of human central nervous system (CNS) is complex. Electroencephalogram (EEG) is one tool to explore the physiological function of brain. It is widely used to diagnose clinical CNS syndromes, such as epilepsy, brain tumor, Parkinson’s disease, etc. However, mostly commercial EEG instruments belong to multi-channel, high price and too big for portable usage. Hence, our thesis tries to design a low price, small volume and single channel of EEG device for easy portable usage. It can use for medical measurements .We used the fourier transform analysis for five volunteers at each time domain in the awake and sleep EEG signals. We found significant differences in sleep and awake waves form according to the volunteers’ sleep and awake statuses. The successful results are given us to have the confidence to test this EEG device in the clinical trials in the near future.

碩士
國立中山大學
機械與機電工程學系研究所
97
One of the important measures for sleep quailty is sleep structure. Normal sleep consists of awake, rapid eye movement (REM) sleep and nonrapid eye movement (NREM) sleep states. NREM sleep can be further classified into stage 1, stage 2 and slow wave sleep (SWS). These stages can be analyzed quantitatively from various electrical signals such as the electroencephalogram (EEG), electro-oculogram (EOG), and electromyogram (EMG). The goal of this research is to develop a simple four-stage process to classify sleep into wake, REM, stage 1, stage 2 and SWS by using a single EEG channel. By applying the proposed approach to 48727 distinct epochs which are acquired from 62 persons, the experimental results show that the proposed method is achieves 76.98% of accuracy. The sensitivity and PPV for wake are 85.96% and 68.35%. Furthermore, the sensitivity and PPV for REM are 82.13% and 74.11%, respectively. The sensitivity and PPV for the stage 1 are 9.02% and 39.00%. The sensitivity and PPV for the stage 2 are 84.19% and 79.36%. The sensitivity and PPV for SWS are 81.53% and 85.40%.

碩士
國立陽明大學
醫學工程研究所
97
Sleep quality refers both the subjective assessment given by the subject of how restorative and undisturbed his/her sleep has been and to a series of objective measures which may be derived from polygraphic recordings. There are many factors could affect the sleep quality like personal behaviors, personal sensation, environment factors, and physiological changes. However, the most commonly used objective measure of sleep quality is an index of sleep fragmentation which may be derived from all three types of recordings. To be compared, object measures of sleep quality cost high and had not necessarily concordant results with subject ones. The aim of this study was to set up a new household sleep quality assessing system. This system was bases on one channel EEG signal to estimate the subjective measure of sleep quality. According to the signal recording, we could define the features of slow wave, alpha wave and spindles in time scale by using zero crossing method, Schimicek algorithm and RMS-amplitude for sleep staging. We used the one channel EEG data (Fpz-Cz) which was downloaded from Europe Data Format (EDF) sleep database to test the algorithm in our system. However, the results of our system had 81.67% accuracy from database hypnogram. Furthermore, we assessed more clinical sleep parameters, including total sleep time, sleep latency and percentage of light sleep, REM sleep and deep sleep to have an evaluation of household system by using sleeping quality score. . The sleeping quality score in normal human is 2.63±1.30 (n=8) evaluating form EDF database and the score in the patients with sleep syndrome is -0.15±2.54 (n=13) evaluating form MIT-BIH database. It was indeed different between these two groups. Therefore, 24 sets of EEG signal from 21 participators (29.13±9.47 years old, 15 males/ 6 females) were used in our experiments and separated into 2 groups by objective sleep stage. One was participator with good-sleep quality and another was participator with bad-sleep quality group. The average score of good-sleep group was 2.61±1.29 and these of bad-sleep group was -0.5±1.22. Therefore, our system could provide an effective index to assess the evaluation of sleep quality. According this study, we set up a household sleep quality system and help people to determine their sleep problem primarily and hope could raise living quality of human.

碩士
國立中山大學
機械與機電工程學系研究所
96
One of the important topics in sleep medicine is sleep structure. Normal sleep consists of rapid eye movement (REM) sleep and nonrapid eye movement (NRME) sleep states. NREM sleep can be further classified into stage 1, 2 and slow wave sleep (SWS) according to the current sleep scoring standard. Among them, SWS has been considered to be very important due to its r restorative value. The goal of this research is to detect SWS by using a single channel EEG signal. Its applications can be divided into two phases. In the first phase, a personalized SWS detector is designed for each individuals By combining these personalized SWS detectors, the second phase develops a general SWS detection method that can be applied to general population with any personalized training process. By applying the proposed method to 62 persons, the experimental results show that the proposed method, in average, achieves 90.69% classification accuracy 90.09% sensitivity and 93.97% specificity. Our experimental results also demonstrate, when applied to persons with higher AHI (apnoea-hypopnea index) values, the proposed method can still provided satisfactory results.

碩士
國立交通大學
電控工程研究所
100
This research conceives a new way of self-rehabilitation which is more helpful for patients, combined with brain waves to approach a new neuro-rehabilitation system. When the user is imaging hand action, the brain will produce a special kind of brain wave, and the system is recognizing the brain wave to control machine to take user’s hand moving, with the principle of neuro-rehabilitation and the enhancement of the bio-feedback method, then this system can make a better rehabilitating effect. The study analyzes the subjects’ event related motor imagery (MI) by using independent component analysis (ICA) and many other brain wave analyzing methods, such as event related spectral perturbation (ERSP), to find the best frequencies and time to recognize, using the sensor to receive brain waves and DSP chip with some algorithms to control the motor to take targets’ hand moving. Using this method of brain wave analyzing and system architecture, the system can recognize motor imagery brain wave successfully and control the rehabilitation machine quickly, to imitate the self-behaviors of paralyzed or stroke patients. This system is hoped to shorten the rehabilitating time greatly and promote the effect, and even saving many human resources.

碩士
國立清華大學
資訊工程學系所
106
This thesis proposes a single-channel EEG and machine-learning classification for drowsiness detection. Unlike previous systems that are based on multi-channel EEG or other sensing modalities, single-channel EEG is much simpler and more cost-effective, but its accuracy has not been assessed. Binary classification for asleep vs. awake state is well understood, but drowsy state overlaps both and has not been easy to classify accurately. To address this problem, we propose a two-level, semi-supervised classification method to effectively distinguish these three states. Experimental results show our approach to be able to accurately detect drowsiness.

Automated neurocognitive performance assessment (NCP) of a subject is a pertinent theme in neurological and medical studies. NCP signifies the human mental/cognitive ability to perform any allocated job. It is hard to establish any certain methodology for research since the NCP switches the subject in an unknown manner. Sleep is a neurocognitive performance that varies in time and can be used to learn new NCP techniques. A detailed electroencephalographic signals (EEG) study and understanding of human sleep are important for a proper NCP assessment. However, sleep deprivation can cause prominent cognitive risks while carrying out activities like driving, and can even lead to lack of concentration in individuals. Controlling a generic unit in non-rapid eye movement (NREM), which is the first phase of sleep or stage N1is highly important in NCP study. Our method is built on RNN-LSTM which classifies different sleep stages using raw EEG single-channel which is obtained from the openly available sleep-EDF dataset. The single raw channel helps classify the REM stage particularly, because a single raw channel, human motion, and movement are not considered. The features selected constituted as the RNNs network inputs. The goal of this work is to efficiently classify the performance in sleep stage N1, as well as improvement in the subsequent stages of sleep.

碩士
南台科技大學
電機工程系
98
Sleep is an essential and crucial event in the lives of every human being. However due to recent advancement in society it also brought in a new tide of disease in the form of sleep disorders. Therefore, looking into sleep is one purpose of this study. The research aims to develop an automated LabVIEW-based sleep stage classification and monitoring system with a user friendly graphic user interface (GUI) to help doctors save time in classifying sleep stages. Patients on the other hand get a picture of their condition through viewing the interface and to help them know how well they slept the night of the test. In addition, the research will be using only a single parameter oriented recording technology in the form of single channel EEG to do the sleep stage classification. Compared to the traditional Polysomnograph it will reduce the discomfort experienced by the patients every time they undergo sleep examination. The research used power spectrum analysis, sample entropy and wavelet transform as a preliminary step for sleep stage classification. These three parameters can show indications for each respective sleep stage but are not very efficient to be used for sleep stage classification individually. Moreover, SVM has good dependence on data and it requires fewer parameters to tune compared to the conventional neural networks. Thus, the data taken from these three analyses are used corporately as instances to a support vector machine (SVM) classification algorithm to perform sleep stage classification. For the system, the GUI platform can present single or multiple hypnogram/s and several key sleep indices that can be beneficial to doctors and patients in understanding certain sleep conditions. Based from the data taken from Physiobank, the developed system was trained using randomly selected epochs from the combined EEG database. Each EEG database are then tested individually and showed an average accuracy of 83.27% for 8 sets of database. In addition, individual results in classifying wake, light sleep, slow wave sleep and rapid eye movement stages are 85.04%, 65.34%, 93.33%, and 92.41% respectively. In the future, the project can still be tuned by refining the input instances to the classification algorithm and having more test data to improve accuracy in classifying sleep stage.

Electroencephalogram (EEG) is an essential tool used to analyze the activities effectively and different states of the brain. Drowsiness is a short period state of the brain that is also called an inattentiveness state. Drowsiness can be observed during the transition from being awake state to a sleepy state. Drowsiness reduces a person’s attention that increases accidental risks when involved in their personal and professional activities like vehicle driving, operating a crane, working with heavy machineries such as mine blasts. Drowsiness Detection (DD) has a significant role in preventing the problems mentioned above. So many traditional algorithms are proposed to detect drowsiness, but among these, the combination of neuroscience with artificial intelligence can effectively diagnose the state of drowsiness. Neuroscience with artificial intelligence algorithms used to detect drowsiness is also popularly known as brain-computer interface (BCI) systems. Single-channel EEG BCIs are highly preferred for convenient use in real-time applications, even though there are many challenges in the actual experimental process. They are feature extraction, feature selection and choosing the best channel. These challenges have badly affected the performance of the BCI in the detection of drowsiness. In this work, a novel channel selection approach is proposed for a single-channel EEG BCI system by integrating the statistical characteristics of the available channels EEG signal to detect drowsiness state successfully. This thesis addresses some EEG sub-band extraction methods and their limitations. These limitations and practical issues are overcome by proposing a time-domain sub-band based feature extraction procedure using the wavelet packet transformation method. This thesis also addresses the asymmetric feature interference between the subjects. These limitations are overcome by proposing a novel feature selection technique using a nonparametric statistical test. In addition to this, a novel single-channel EEG signal analysis approach and single feature computation are also offered to deploy most quickly on low computing capacity systems. Apart from the machine learning methods, this thesis also discusses a novel deep learning architecture based on a convolutional neural network (CNN) for automated single-channel EEG signal classification to detect drowsiness Subject wise, cross-subject wise, and combined subject’s wise validations are also employed to improve the generalization capability of the proposed techniques in this thesis. The whole work is carried out over prerecorded EEG databases such as Physionet real-time sleep-analysis-EEG and simulated-virtual-driving-driver-EEG.

碩士
國立成功大學
電機工程學系碩博士班
101
Abstract An automatic evaluation system of IADL (Instrumental Activities of Daily Living) is proposed, the system separates the total IADL scores into three categories: high (disability-free, IADL scores from 16 to 24 points), medium (mild disability, IADL scores from 8 to 15 points) and low (severe disability, IADL scores from 0 to 7 points). Single channel EEG device is applied to thirty seniors (from age 70 to 96) of the IADL scores uniform distribution to do the following IADL scenarios: (1) telephone using, (2) financial management. The brainwave data of chatting scenario is collected additionally and 5 features to classify the group of IADL scores are used as follows: (1) Average Amplitude (2) Power Ratio (3) Spectral Centroid (4) Spectral Edge Frequency 25% (5) Spectral Edge Frequency 50%. Besides, LDA (Linear Discriminant Analysis) algorithm is combined with 5 features mentioned above to evaluate IADL score. To find out the best classifier of IADL assessment, not only LDA classifier but SVM (Support Vector Machine) and KNN (K-th Nearest Neighbor) are used to compare the accuracy of IADL evaluation, and LOOCV (Leave-One-Out Cross-Validation) is used to verify the proposed system. Finally, the accuracy is about 90% and also higher than the other two classifiers when using LDA under the same feature. The groups of IADL scores of patients are classified exactly when using the brainwave data of chatting scenario. The proposed system can help doctors to evaluate the results objectively before proceed the IADL interview to patients and combine with the judgment of doctors, the objective and accurate IADL scores are obtained.

碩士
國立高雄第一科技大學
系統資訊與控制研究所
101
Eyes are the window of the soul which use for receive the external messages. However, some kinds of patients that eyes can rotate but the limbs are paralyzed calling Spinocerebellar Ataxia(SCA). On the other hand, the Amyotrophic Lateral Sclerosis (ALS) can receive the external information through hearing and sight sensorial channels but can’t reply to doctor or family by talking or body language. So we try to design easy operation and have high compatibility devices for communication. The purpose of this study is control the mechanism of eye movement commands with a single channel EEG headset. It can catch the eye movement signals through the single channel wireless EEG headset, then analyze and translate to multi-commands by the programming language. The commands of detecting by the eye movements with up, down, right, left, and voluntary winks based on approaches such as the set threshold algorithms, four-quadrant characteristics algorithms and eight-quadrant characteristics algorithms. Finally, the Chinese text-messaging based on Braille input via Brain-Computer Interface(BCI) are shown and pronounced out in the android phones.

碩士
國立臺灣大學
生醫電子與資訊學研究所
104
People pay attention to their sleep quality and sleep problems. When people don''t have enough or qualified sleep, these conditions may have negative impacts on people''s life and their efficiency in work. In order to solve these problems, many hospitals set up sleep quality centers where professional instruments and consultations are used to solve sleep problems, but these advanced instruments and human efforts cost a lot. Besides, people must be stuck with many electrodes to collect signals, such as electroencephalogram (EEG), electrocardiogram (ECG), electromyography(EMG), oxygen desaturation index and so on, before people receive their diagnosis and analysis for their sleep conditions. During the examination, people with electrodes feel uncomfortable due to lots of electrodes influencing their sleep, so the results might be incorrect and unable to reflect the real conditions. To address these problems, we propose the Takagi-Sugeno fuzzy neural network-based algorithm with single-channel EEG signal for the discrimination between light and deep sleep stage. This main algorithm is using the single-channel EEG to combine signal processing and Takagi-Sugeno neural network to discriminate between light and deep sleep. The advantage of using the single-channel EEG is decreasing people''s uncomfortable feeling, reflecting the real sleep conditions, and increasing the accuracy by using two electrodes to get the EEG signal.

博士
國立臺北科技大學
電子工程系
107
This dissertation proposes to recognize a user’s intentions in selecting from a set of machine-controlling commands by measuring his/her brainwaves captured with single channel electroencephalogram (EEG). Our strategy is to convert a multiple-choice decision into yes-no decisions. For example, in a task of dialing assistance, our system prompts the user to select from each of the digits, and then analyzes his/her brainwave to determine if the prompted digit is what he/she wants. Assume that the user’s intention is 7. Then, when the system prompts the user whether to choose digit 7, the resulting EEG measured from the user should present a certain pattern of “Yes”; otherwise, the result should present a certain pattern of “No”. Hence, our system's goal is to determine whether the user’s intention is “Yes” or “No” based on the measured EEG. This dissertation uses a simple, portable, and cheap instrument that extracts a single-channel EEG from the user’s frontal lobe. The underlying beta waves of EEG are then distilled and recognized to determine the user’s intention. We compare the four recognition methods, respectively, based on Gaussian Mixture Models (GMMs), Hidden Markov Models (HMMs), Multi-Layer Perceptron (MLP), and Recurrent Neural Network (RNN), and conducts experiments using 2400 test EEG samples recorded from 10 subjects. And the recognition accuracy obtained with our system is 79.2%

The popular technique used for detection of fetal heart rate before delivery is Fetal Electrocardiogram (FECG). It shows the muscular function and electrical activity of the fetus heart. It represents the characteristics such as dynamic behaviors, waveform and heart rate of the fetus. These characteristics help to determine the fetal development, the existence of fetal distress, fetal life, fetal maturity or congenital heart disease. These above characteristics help the doctors for appropriate treatment during pregnancy. The heart rate of the fetus can easily be detected after estimation of the fetal ECG signal from the abdominal ECG signal. The abdominal ECG signal is collected by placing electrode at the abdomen area of the mother. The abdominal ECG signal contains fetal ECG signal, maternal ECG component, and noise. To estimate the fetal ECG signal from the abdominal ECG signal, removal of the noise and the maternal ECG component presented in it is very much necessary. Maternal ECG component is the dominant part of the abdominal ECG signal. To remove the maternal ECG component present in the abdominal ECG signal, accurate detection of the maternal R peaks from the abdominal ECG signal is required. An efficient R peak detection technique is required for the detection of accurate maternal R peaks from the abdominal ECG signal and the accurate detection of R peaks of the extracted fetal ECG signal as well. So that heart rate of the fetus can be computed. However, almost all existing R peak detectors suffer due to the non-stationary behavior of both QRS morphology and noise. To overcome these difficulties, we have proposed a three-stage improved method to detect R peaks based on Shannon energy envelope. The proposed R peak detection method in this dissertation shows improved performance compared to other existing methods available in the literature. In the recent years, Extended Kalman Smoother (EKS) has been used and has shown good performance for extraction of the fetal ECG signal from the single channel abdominal ECG signal. But the limitation of this method is that it fails to extract fetal QRS complex if it is overlapped by the maternal QRS complex in the abdominal ECG signal. The method also sometimes requires operator’s interaction to initialize the parameter of EKS to extract the fetal ECG signal which is dependent on abdominal ECG signal for better performance. Author of this thesis has investigated the effectiveness of Adaptive Neuro-Fuzzy Inference System (ANFIS) with EKS for extraction of the fetal ECG signal using single channel abdominal ECG signal. The EKS with ANFIS method proposed in this work for fetal ECG extraction is found to detect fetal QRS complex even if it is overlapped by the maternal QRS complex in the abdominal ECG signal. In the EKS with ANFIS framework, proposed Shannon energy based R peak detection is used for detection of the maternal R peaks from the abdominal ECG signal. The EKS filtering framework for denoising purpose requires operator’s interaction. To avoid the operator’s interaction and also to provide better performance using EKS framework, author has investigated the effectiveness of the Extended Kalman Smoother (EKS) with the Differential Evolution (DE) technique for noise cancellation of the ECG signal. DE is used as an automatic parameter selection method for the selection of 10 optimized parameters of the ECG signal, and these are used to create the ECG signal according to the real ECG signal. Also, these parameters are used in the EKS algorithm for the development of the state equation and initialization of the parameters of the EKS. The EKS framework is used for denoising of the ECG signal from the single channel recording. The effectiveness of the proposed noise cancellation technique has been evaluated by adding white, colored Gaussian noise and real muscle artifact noise at different SNR to some visually clean ECG signals from the MIT-BIH arrhythmia database. The proposed noise cancellation technique of ECG signal shows better Signal to Noise Ratio (SNR) improvement, lesser Mean Square Error (MSE) and Percent of Root mean square Distortion (PRD) compared to other well-known methods. Finally, the author has proposed a five-stage based methodology for further improvement of extracted FECG from the single channel abdominal ECG using DE algorithm, EKS and ANFIS framework. The pre-processing stage is used to remove the noise from the abdominal ECG signal and the EKS framework is used to estimate the maternal ECG signal from the abdominal ECG signal. The optimized parameters of the maternal ECG component (signal) are required to develop the state and measurement equation of the EKS framework and the same are selected by the DE algorithm. The relationship between the maternal ECG signal and the available maternal ECG component in the abdominal ECG signal is nonlinear. To estimate the actual maternal ECG component present in the abdominal ECG signal and also to recognize this nonlinear relationship, the ANFIS is used. Inputs to the ANFIS framework are output of the EKS and the pre-processed abdominal ECG signal. The fetal ECG signal is computed by subtracting output of the ANFIS from the pre-processed abdominal ECG signal. Non-invasive fetal ECG database and set A of 2013 physionet/computing in cardiology challenge database (PCDB) are used for validation of the proposed methodology. This thesis also describes a Field Programmable Gate Array (FPGA) implementation of a heart rate calculation system using Electrocardiogram (ECG) signal. The proposed FPGA based heart calculation system is FPGA implementation of proposed R peak detection technique based on Shannon energy envelope with a little modification. After heart rate calculation, tachycardia, bradycardia or normal heart rate can easily be detected. Heart rate is calculated by detecting the R peaks from the ECG signal. To provide a portable and the continuous heart rate monitoring system for patients needs a dedicated hardware. FPGA provides easy testability, allows faster implementation and verification option for implementing a new design. We have proposed a five-stage based methodology by using basic VHDL blocks like addition, multiplication and data conversion (real to the fixed point and vice-versa) etc for our proposed design. The proposed FPGA based heart rate calculation (R-peak detection) method shows better performance compared to other well-known methods for detection of R peaks (heart rate calculation) and successfully implemented in FPGA.

碩士
中原大學
電機工程研究所
94
Health of fetal is the most important concern of pregnant women and gynecologist, and fetal ECG (F-ECG) is an index to determine the health of fetal. In general, using non-invent fetal electrocardiogram(FECG) monitoring system to observe fetal ECG is very convenient. However signals obtained from maternal abdomen are contaminated by maternal ECG (M-ECG), and interference from electric device. Therefore, if we want to obtain fetal ECG, we have to remove these non-demand signals. In order to remove the non-demand signals, our proposed method exploits this feature for selective separation of M-ECG and F-ECG components by formulating the problem in the singular value decomposition (SVD) framework. In this paper, in order to obtain QRS wavelets we used filter to remove the interference low-frequency trend component and reduce interference of M-ECG’s T wavelet. Then we used the Singular Value Ratio (SVR) spectrum, developed on the basis of the SVD, to detect the periodic components. SVR spectrum can provide an estimate of the period length of the most dominant periodic component present in any signal. Follow we according to the periodic length obtain a spectrogram matrix. Final we used Short Time Fourier Transform (STFT) and SVD to separate F-ECG.

碩士
國立交通大學
生醫工程研究所
105
Sudden deaths occurring during exercise or during a period after exercise is unexpected. Previous studies have shown that approximately 90% of these sudden deaths are due to heart problems. Some cardiovascular abnormalities occur only in the state-of-motion and the ECG changes measured in this state is termed as exercise-ECG, which is also a measurement for a stress test. However, ECG signal recording is affected by external or internal effects, such as wire movement, baseline drift, EMG signal, etc. Apart from R-R interval changes, occurrence of ST-segment depression is an indicative of the presence of obstructive coronary artery disease. Low frequency signal component, the PQST wave, is also known to be a promising indicator of heart condition. Therefore, capturing accurate ECG waveform and removing artifacts is necessary. There are many approaches for artifact removal with each method having its own limitations, and only few studies address the artifact issues for the exercise-ECG analysis. Therefore, the main aim of this study is to develop a novel method of using canonical correlation analysis to remove high and low-frequency noise in the ECG. In addition, the advantage of the proposed technique is that it can be applicable to single-channel system. The results showed that, in the resting state, the signal-to-error ratio of this method at least one time higher than the traditionally adopted methods; the root-mean-square error at least 0.1 ~ 0.9 mV less in the motion state; The proposed method also displayed outstanding performance in terms of computing speed. This study has the potential for developing a commercial real-time system in home care and medical examination.

碩士
國立陽明大學
醫學工程研究所
98
Electroencephalogram analysis is usually used as a clinical method to monitor sleeping status, but the method is complex and time-consuming. Furthermore, electroencephalogram measurement needs to arrange electrodes with fixed distance, which requires professionals' help and unsuitable for in-home utility. It is then necessary to develop a sleeping evaluation system for home use. The relationship between autonomic nervous system and heart rate variability is an interesting issue. This study then proposed a method based on electrocardiogram to analyze sleeping stages and to establish a home use evaluation system for sleeping quality. With the R&K rule, sleeping stages is divided to three types. They are wake and rapid eye movement period (WR), CAP and Non-CAP. In this study, 7 normal, 2 mild and 1 moderate sleep apnea patients from Shin Kong Hospital were recruited as the study subject. All subjects were monitored with the records of electroencephalogram, electrooculogram and electrocardiogram. Electroencephalogram and electrooculogram were recorded by using polysomnographic apparatus, while electrocardiogram was measured by using the real-time physiological monitoring system (VitalTag WHC-BT-03) of Aescu Technology, Inc. The results were interpreted by hospital professionals, and electrocardiogram was analyzed using a Matlab program. The major analysis was on non-dynamic heart rate variability, including Poincare Plots of SD1/SD2 and elliptical area (S). Moreover, Support Vector Machine is trained to classify sleeping stages. The training results showed that SD1/SD2 &lt;0.36 was judged to be WR stage, 0.36&lt; SD1/SD2 &lt;0.61 to be CAP stage, and SD1/SD2> 0.61 to be Non-CAP stage. Three additional subject signals were used to evaluate the system, and its overall accuracy was 72.67%.