본문 바로가기 주메뉴 바로가기
22대 대한민국 국회 국회정보길라잡이 국회의원검색
회원가입 로그인
HOME

정보검색

소장정보 검색

국회도서관 홈으로 정보검색 소장정보 검색
결과 내 검색 동의어 포함
결과 내 검색 동의어 포함

-

목차보기

Title Page

Acknowledgements

Contents

I. Introduction 9

II. EEG and Epilepsy 15

2.1. EEG artifacts 19

2.2. EEG in epilepsy diagnosis 22

2.3. The available patients’epileptic data 27

III. Continuous wavelet transforms for seizure detection 31

3.1. Wavelets for EEG signal analysis 33

3.1.1. Research methodology 35

3.1.2. Proposed wavelet functions 36

3.2. Theoretical formulations 39

3.2.1. Scale setting 42

3.2.2. Wavelet thresholding technique 45

IV. Seizure classification using artificial neural networks 55

4.1. Artificial neural networks 56

4.2. Computational models of neurons 58

4.3. Backpropagation learning rule 60

4.4. Neural network methodology 63

V. Experimental results and discussions 68

5.1. Tonic clonic seizure detection 68

5.2. Temporal lobe seizure detection 85

5.3. EEG seizure classification 105

VI. Conclusion 115

References 118

Abstract 124

요약 126

List of Tables

Table 2.1. Clinical data of the six patients studied in this research 27

Table 5.1. Detection accuracy at dyadic scale combinations using wavelet functions db2, db5, bior1.3 and bior1.5. 103

Table 5.2. Performance of the various ANN architectures 109

Table 5.3. The offline validation results for the proposed algorithm which is applied to patients data sets 111

Table 5.4 The GDR performance results for time varying EEG signal as the input segments to the ANN 113

List of Figures

Fig. 2.1. The 10-20 international system of electrode placement. 17

Fig. 2.2. Example of electrode cap with electrodes placed on a subject’s scalp using 10-20 electrode placement system. 18

Fig. 2.3. A segment of 8 channel EEG signal, a) noise free EEG and b) with EMG noise caused by body movement. 20

Fig. 2.4. The EEG noise tree physiological and non-physiological sources. 21

Fig. 2.5. Examples of EEG transient events, a) spike, b) spike and slow wave, c) polyspikes and d) sharp wave. 23

Fig. 2.6. EEG signal samples obtained from the five channel recordings for representation of varying signal waveforms. 24

Fig. 2.7. A taxonomy of seizures based upon classifications from the International League Against Epilepsy (1985). 26

Fig. 2.8. Three types of EEG signals studied in this research, a) Normal, b) Inter-ictal and c) Ictal segments. 29

Fig. 2.9. The thirty three channel recording of EEG from epileptic patients and monitoring during the seizure evaluation. 30

Fig. 3.1. Research methodology, a) data selection, b) continuous wavelet transforms of EEG and c) three layer feedforward neural networks. 35

Fig. 3.2. Four different mother wavelet functions, a) bior1.3, b) bior1.5, c) db2 and d) db5. 36

Fig. 3.3. The scaling and shifting process of wavelet functions along with the analyzed time series. 38

Fig. 3.4. A wavelet thresholding example of a noisy EEG signal, a) a segment of EEG and b) the double thresholded wavelet coefficients. 54

Fig. 4.1. McCulloch and Pitt’s neuron model. 58

Fig. 4.2. Different types of activation functions. 59

Fig. 4.3. A typical multilayer feedforward network architecture. 60

Fig. 4.4. The three layer feedforward neural network structure with seven inputs, variable number of hidden neurons and one output. 63

Fig. 5.1. EEG signal corresponding to a tonic-clonic seizure of an dpileptic patient KW, the seizure starts at 80 second time and ends around 160 seconds of time segment. 69

Fig. 5.2. Preseizure EEG signal, a) the analyzed signal segment and b) wavelet representation the signal locations of the detected epileptic transients by using bior1.5 wavelets. 72

Fig. 5.3. Two scale wavelet decomposition of full EEG signal of Tonic Clonic seizure using db2 wavelet function. 73

Fig. 5.4. Contour map view of two scale wavelet decomposition of Fig. 5.3 using db2 wavelet function at scales {a6} and the detected epileptic transients are localized at bottom plane of the figure. 74

Fig. 5.5. Combination of wavelet coefficients from five scales {a2, a4, a6, a8, a10} of the tonic clonic EEG singal which is represented using a) bior1.3 b)bior1.5 functions. 75

Fig. 5.6. Analysis of full EEG, a) EEG signal of Tonic Clonic seizure from b) contour map image and c) pure extracted epileptiforms. 76

Fig. 5.7. Details of epileptic EEG signal, a) the process of wavelet detection using db5 and its correlation with signal from b) {a6} scale and c) {a8} scale. 78

Fig. 5.8. EEG signal corresponding to another tonic-clonic seizure of an epileptic patient SH, the seizure starts at 65 second time and ends around 125 seconds of time segment. 80

Fig. 5.9. Detection of epileptic spikes at four scales using db5 wavelet function, a) localized wavelet coefficients superimposed with the background signal b) is the contour map image of wavelet cofficients at corresponding scales as well as the time located spikes is given at the bottom. 81

Fig. 5.10. Detection of epileptic spikes at four scales using bior1.5 wavelet function, a) localized wavelet coefficients superimposed with the background singal b) is the contour map image of wavelet coefficients at corresponding scales as well as the time located spikes is given at the bottom. 82

Fig. 5.11. Details of EEG epileptiform detection, a) wavelet function’s capturing the epileptiforms and the thresholding process with bior1.3 function at b) {a6} scale and c) at {a8} scale. 83

Fig. 5.12. Three types of the EEG signal of temporal lobe epileptic data, a) normal, b) inter-ictal and c) ictal segments. 86

Fig. 5.13. Wavelet decomposition of seizure signal using db5, a) EEG segment, b) time-scale representation. 87

Fig. 5.14. Seizure EEG signal and wavelet decomposition using db.5 at scales {a2, a4, a6, a8}. 89

Fig. 5.15. Analysis of seizure EEG segment, a) short seizure evolution b) detection of onset and offset points and c) time localization of epileptiforms. 90

Fig. 5.16. Details of a preseizure period of EEG signal, a) 5.5 second EEG segment, b) the wavelet coefficients of signal using db5 function. 91

Fig. 5.17. Ictal EEG segment superimposed with detected epileptiforms, by using a) db5 and b) bior.15 wavelets at scales {a6, a8}. 93

Fig. 5.18. Interictal EEG segment superimposed with detected epileptiforms by using bior1.5 wavelets at b) {a6} and c) {a8} scales. 94

Fig. 5.19. A combined multiscale representation of ictal EEG transients a) analyzed ictal EEG segment, wavelet decomposition using b) db2 and c) db5. 96

Fig. 5.20. A combined multiscale representation of ictal EEG transients, a) analyzed ictal EEG segment, wavelet decomposition using b) bior1.3 and c) bior1.5. 97

Fig. 5.21. Interictal EEG transient and its multiscale representation, a) analyzed interictal EEG segment, wavelet decomposition b) db2 and c) db5. 98

Fig. 5.22. Detection of inter-ictal EEG transients, a) EEG signal and using bior1.3 and c) bior1.5 from combined multiscale representations. 99

Fig. 5.23. EEG inter-ictal signal wavelet decomposition using bior1.3 function a) signal is superimposed with noisy wavelet coefficients at {a4} scale , b) {a6} and c) {a8} accordingly. 101

Fig. 5.24. EEG inter-ictal signal and pure wavelet thresholded coefficients using bior1.3 function a) signal is superimposed with wavelet thresholded coefficients at {a4} scale , b) {a6} and c) {a8}. 102

Fig. 5.25. Wavelet input features for neural network classifier as positive, negative wavelet amplitude and duration of wavelet wave coefficients. 106

Fig. 5.26. Accuracy results of ANNs from dyadic scale combinations using wavelet functions. 107

Fig. 5.27. Performance of neural network classifier from two error minimizing algorithm, a) gradient decent, b) scaled gradient conjugate. 110

초록보기

발작신호의 기록은 간질 신호의 과도전류 값을 구하는데 가장 큰 관심이 있다. 발작은 뇌의 일부 또는 뇌 전체에서 일어나는 리듬 현상으로, 그 발작증세가 개인에 다라 수초에서 수분간 지속된다. 일반적으로 발작은 드물고 예측이 불가능하게 발생한다. 뇌파가 장시간 저장되는 동안, 발작 형상이 자동적으로 검출되는 것은 매우 매력적이다. 뇌파 신호는 변화가 많기 때문에 주파수 분석 같은 일반적인 방법들로는 지단의 목적을 달성할수 없다. 본 논문에서는 뇌과속에 혼재되어 있는 여러신호들과 함께 저장되어 있는 간질 과도 전류 값을 검출하고 분류하기 위해 연속 웨이블렛 변화과 신경망을 이용한 새로운 방법을 제안한다. 제안된 방법은 데이터 취득, 특징 추출 그리고 분류 단계로 구성된다. 특징 추출 단계에서는 최적의 기본 웨이블렛 기저 함수와 웨이블렛 임계치 방법을 사용한다. 이 단계에서 제안된 방법은 발작 활동 상태의 시점의 구분하고 정확하게 알아내는데 매우 효과적일 뿐만 아리나 사전-발작, 발작 그리고 사후-발작과 같은 발작 시작 시점을 구분하고 정확하게 알아내는데 매우 효과적임을 입증한다.

분류 단계에서는 역전파 학습 공식에 따라 다층 퍼셉트론 신경망을 사용한다. 분류기의 입력 계수들은 발작형태와 일치하는 웨이블렛 계수들로 구성된다. 본 논문에서는 두 가지 다른 훈련 알고리듬과 높은 분류 정밀도를 갖는 교차 검증 결과를 통하여 성능을 평가하였다. 분류 단계에서 SVM(support vector machine) 알고리즘 지원되다면 이 연구에서 제안된 시스템의 정확도는 더욱 향상될 것이다.

추천서가 (다양한 추천 자료를 만나보세요)

권호기사보기

권호기사 목록 테이블로 기사명, 저자명, 페이지, 원문, 기사목차 순으로 되어있습니다.
기사명 저자명 페이지 원문 기사목차