國立中山大學學位論文典藏

Transcription

1 立 論 度 數 A Digital Oximetry Based Method for Estimating Respiratory Disturbance Index 年

2

3 ..I III Data Base SpO 2 Data base Vázquezz Magalang I

4 Coefficient of Skewness Skewness Coefficient of Skewness Coefficient of Skewness Skewness MLP MLP Averaging Committee II

5 1.1 OSA CPAP machine Polysomnography RDI Autocorr Digital Pulse Oximeter pulse oximetry Alice GRASS Alice RDI [Vázquezz et al.,1999] respiratory disturbance RDI ( data base) AHI AHI [Magalang et al.] AHI AHI III

6 database OSA even (RDI:100.8) OSA even (RDI:0) RD RD Skewness Skewness Skewness Skewness V Histogram (RDI:0) VHistogram(RDI:100.8) SkewnessRDI RDI MLP MLP Averaging Committee Averaging Committee 44 IV

7 Accuracy ( ).. 50 Skewness Skewness ( ).. 51 Skewness. 52 Skewness ( ) V feature V feature ( ).. 53 V feature V feature ( ) SkewnessV feature Skewness V feature ( ) V feature V feature ( ) ( ) V

8 5.21 ( ) Skewness Skewness ( ) % RDI % RDI % RDI Skewness Skewness Skewness VI

9 5.39 V feature V feature V feature VII

10 2.1 Data base RDI Calculations of Accuracy VIII

11 SAS 90 OSA OSA RDI IX

12 A Digital Oximetry Based Method for Estimating Respiratory Disturbance Index Abstract SAS has become an increasingly important public-health problem in recent years. It can abversely affect neurocognitive, cardiovascular, respiratory diseases and can also cause behavior disorder. Moreover, up to 9 of these cases are obstructive sleep apnea (OSA). Therefore, it is important that how to diagnose, detect and treat OSA. The respiratory disturbance index is one parameter of estimating OSA. Polysomnography can monitor the OSA with relatively fewer invasive techniques. However, polysomnography-based sleep studies are expensive and time-consuming because they require overnight evaluation in sleep laboratories with dedicated systems and attending personnel. Based on the digital oximetry, this work introduces the estimating respiratory disturbance index. In particular, via signal processing, feature parameters and artificial intelligence, this thesis describes an off-line SpO 2 -based RDI estimating system. X

13 Guilleminut Dement obstructive sleep apnea syndrome, OSAS1979 Block hypopneas OSA CSA [Chazal de et al., 2003] 4% 2% [Young et al., 1993; Penzel et al., 2000; Atul and David, 2002] [Young et al., 1993] 1.1 1

14 [Behbehani et al., 1995; Atul and David, 2002][ 2004] SAS [Lavie et al., 2000; Paola and Virend, 2001; Atul and David, 2002] [Teran et al., 1999; Atul and David, 2002] [Young et al., 1997] [ 2004] 2

15 1.1 OSA 3

16 OSA 1.2 OSA Continuous Positive Airway Pressure Machine, CPAP Machine [Atul and David, 2002] 95% OSA [Behbehani et al., 1995; Rosenthal et al., 2000; Montserrat et al., 2001] [Behbehani et al., 1993; Behbehani et al., 1995; Yen et al., 1997] 1.2 Apnoea/Hypopnoea IndexAHI Respiratory/ Disturbance IndexRDI [AASM, 1999] 4

17 Polysomnography[AASM, 1999] OSA ; OSA ; OSA OSA [Behbehani et al., 1995; Yen et al., 1997; Montserrat et al., 2001; Michael et al., 2002] OSA OSA 1.2 CPAP machine OSA 1.3 5

18 EEG EMG EOG ECG [Chazal de et al., 2003] [Whitney et al, 1998] [ 2004] 1.3 Polysomnography 6

19 [Young et al., 1997] 5000 [Chazal de et al., 2003] OSA PSGPolysomnography 1.2 RDIRespiratory disturbance index Apnea Hypopnea OSA 7

20 Apnea Hypopnea (oxygen desaturation) (arousal) OSA SpO 2 SpO SpO 2 SpO 2 RDI 1.4 8

21 度 理 數 數 類 神 路 RDI 立 離 RDI 1.4 離 RDI 錄 流 1.4 論 論 六 論 裡 說 了 塞 療 論 了 度 念 data 9

22 base 10

23 2.1 Data Base [, 1998] (oxygenated hemoglobin) (reduced hemoglobin) (HbO 2 ) (Hb) [Kevin S et al., 1991;, 1998] [, 2004] 2.2 SpO 2 97 CO-Oximeter Pulse Oximeter

24 2.1 Autocorr Digital Pulse Oximeter (

25 nm 940nm [, 2004] 2.3 [, 2004]

26 OSA [Ulysses et al., 2003; Shinji et al., 2002; Deegan and McNicholas, 1996 ] BCI-3420($287 ) BCI-3401($625 ) BCI-3303($895 ) BCI-3301($489 ) BCI-3402($805 ) BCI-3403($699 ) ( pulse oximetry 14

27 ( ( SpO 2 Data base Data base Data base polysomnographic system Alice4, Healthdyne Technologies, Atlanta, Georgia, USA 2.6 electroencephalogram, EEG electrooculogram, 15

28 EOG mentalis / submentalis electromyogram, M.S. ECG anterior tibialis electromyogram, A.T. EMG electrocardiogram, ECG nasal and oral airflow chest and abdominal breathing effort blood oxygen saturation, SaO 2 snoring body position nocturnal penile tumescence, NPT esophageal pressure esophageal ph pulmonary artery pressure blood pressure, BP arterial carbon dioxide tension, PCO 2 [, 2004] 2.6 Alice4 [, 2004] 16

29 eledtrode 2.7 sensor 2.8 bio-electrical potential GRASS [, 2004] 2.8 [, 2004] 17

30 Data base Mean(SD) Range Age(years) 43.11(13.57) 2-78 F/M 58/245 Height(cm) (11.32) Weight(kg) 71.83(14.33) Body mass index(/m 2 ) 25.91(4.11) Apnoea hypopnoea index(/h) 36.20(24.52) Data base 18

31 2.7 Alice4 C3A2 C4A1 LEOG REOG MicL ECG RR FLW THO ABD SaO 2 Body LEMG LUX STAGE 19

32 3.1 OSA OSA [Magalang et al., 2003;Valerie et al., 2003; Carlos et al.,2003; Michael et al., 2003; Ulysses et al., 2003; Shinji et al., 2002; Vázquezz et al.,1999;bertrand et al., 1999; Deegan and McNicholas, 1996] RDI[Magalang et al., 2003; Vázquezz et al.,1999] RDI Vázquezz Vázquezz 3.1 RDI RDI 30 respiratory disturbance 1. SpO 2 mark 2. mark 3 SpO SpO 2 20

33 baseline 4 baseline mark respiratory disturbance respiratory disturbance RDIrespiratory disturbance index Vázquezz et al

34 3.1 RDI X AHI-A RDIY RDI [Vázquezz et al.,1999] 3.3 RDI X Actual RDIY Predicted RDI 22

35 V feature Magalang Magalang Vázquezz 1. index index moving window window 12 shift1 window 2. log 10 AHI+1= BF BF2 BF1index BF index 0 0 AHI RDI Magalang et al

36 RDI AHI AHI [Magalang et al., 2003] 3.5 AHI AHI database 24

37 RDI SpO 2 Artifacts SpO 2 SpO 2 Artifacts Artifacts SpO 2 Artifacts SpO

38 Pulse Oximeter SpO 2 SpO peak 93% 26

39 3.8 valley 97% 2 Artifacts Y[k]k k If ( y [k+1] y [k] > y [k+1] 4) and ( y [k+1] y [k+2] > y [k+1] 4) y [k+1] = y [k] If (y [k] y [k+1] > y [k] 4) and (y [k+2] y [k+1] > y [k+2] 4) y [k+1] = y [k] 27

40 OSA OSA OSA even 1 OSA 0 RDI

41 3.10 OSA even 1 OSA 0 RDI 0 RDI moving window window 30 shift 1 window overlapping window V 4. 3 V 5. data base

42 RDI RD X SpO RD X SpO 2 30

43 RDI Mapping skewness RDI Coefficient of Skewness Skewness skewness Skewness 31

44 3.14 Skewness 3.15 Skewness 3.16 Skewness Coefficient of Skewness 32

45 Coefficient of Skewness Coefficient of Skewness Coefficient of skewness 1. Pearson s formula K. Pearson SK p = 3 e (µ M ) σ µ = M e = median s = 1SK p = 0 µ=m e 2SK p 0 µm e µm e 3SK p 0 µm e µm e 2. Bowley s formule 33

46 coefficient of skewness SK B Q = 3 + Q1 2 Q 3 M Q 1 e Q 1 Q 2 Q 3 Q 4 1SK B = 0 Q 3 M e = M e Q 1 2SK B 0 Q 3 M e M e Q 1 3SK B 0 Q 3 M e M e Q 1 4 SK B 0.1 SK B E( x µ ) y = 3 σ 3 µ x s x Et y0 y0 [ 2002] Skewness V Histogram RDI

47 RDI0 RDI V Histogram 3.18 V Histogram RDI SpO2 V Skewness Data base V skewness RDI Skewness RDI 35

48 3.5 4 Skewness 3.1 V feature moving window window 30 shift 1 window overlapping window FFT S 4. 3 V 5. data base 303 FFT 95 RDI

49 RDI(event/hour)) 3.20 RDI RDI Mapping 37

50 Chazal Linear Discriminant ModelLD) Quadratic Discriminant ModelQD Maximum LikelihoodML [Chazal de et al., 2003] 4.1 [Haykin, 1999] [Rossiev et al., 1995; Naguib et al., 1998; Ranta et al., 2002; Robert et al., 2002] [Atiya et al., 1999; Chang et al., 2001] [Wilson and Recknagel, 2001] [Charylonink and Chen, 2000; Hippert et al., 2001; Senjyu et al., 2002] [Pattie and Snyder, 1996] [Cao and Tay, 2001; Gençay and Qi, 2001; Rech, 2002] [Marcelo and Álvaro, 2000] 38

51 4.1.1 MLP Neural Network Model Multilayer PerceptronMLP) MLP Simon Haykin Neural Networks [Haykin, 1999] 4.1 MLP Connection Weight 39

52 Connection Weight Input Signal Output Signal Input layer Hidden layer Output layer 4.1 MLP 4.2 MLP Input Signal Output Signal Mapping Error SignalOutput Layer Hidden Layer Hidden Layer Output Layer Connection Weight Error SignalHidden Layer Input Layer Input Layer Hidden Layer Connection Weight 4.2 MLP MLP Back-Propagation Algorithm [Haykin, 1999] Function Signal Error Signal 4.2 MLP 40

53 4.1.2 MLP Input X Desired Output D N {( x N i. d i )} i= 1 xi d i x Desired d 1 x Desired Output -1 i MLP i Learning Rate MLP x y i i i x Desired Output i d i 1 x i Desired Output 0 MLP 0.5 y 0.5 y 0.5 i i 4.2 (Committee Machine) Nilsson 1965 [Haykin, 1999] (Expert) 41

54 Averaging Committee[Haykin, 1999]Bagging Ensemble [Leo, B., 1996]AdaBoost[Yoav, F., and Robert, E. S., 1997] (1) Bagging Ensemble (2) AdaBoost (1)(2) 4.3 Averaging Committee Averaging Committee Machine [Haykin, 1999] 4.3 Averaging Committee Machine Expert Expert Averaging Committee Machine 42

55 Input x (n) Expert 1 Expert 2 y 1( n) y 2( n) Output y (n) Combine Expert k (n) y k y( n) = 1 k k i= 1 y i ( n) 4.3 Averaging Committee 43

56 論 論 了 參 數 類 來 兩 來 數 來 行 類 類 神 路 類 力 更 RDI 狀 論 5.1 類 神 路 了 10 MLP 5.1 y ( n 1 ) Input x (n) MLP 1 MLP 2 y ( n 2 ) Output y (n) Combine MLP 10 y ( n 10 ) y( n) = i= 1 y i ( n) 5.1 Averaging Committee 44

57 data base training data validation data testing data Early Stop RDI RDI Skewness 5 V feature

58 Input MLP 5 10 Learning Rate Early Stop Type w1*training accuracy + w2*validation accurary Early Stop Epoch 50 Epoch 1000 Averaging Committee w1 = Training. Number/(Training Number + Validation Number) w2 = Validation. Number/(Training Number + Validation Number) 5.2 forward selection 35 0~ Training AccuracyValidation Accuracy Testing Accuracy mean standard deviation 5.2, Training Accuracy Validation Accuracy Testing Accuracy RDI threshold (%) (%) (%)

59 Accuracy Accuracy 5.3 Training accuracyvalidation accuracy Testing accuracy sensitivityspecificitypositive accuracy negative accuracy meanstandard deviation 5.3 Sensitivity Specificity Positive accuracy Negative accuracy (%) (%) (%) (%)

60 (%) (%) (%) (%) (%) correlation coefficient RDI RDI RDI 5.4 1~3 5 48

61 RDI>30 RDI<30 RDI>30 TP FP RDI<30 FN TN 5.7 Calculations of Accuracy 5.7 sensitivityspecificitypositive accuracy negative accuracy data base 158 RDI RDI 30 data base Sensitivity Specificity TP RDI>30 RDI>30 FN RDI>30 RDI30 FP RDI30 RDI30 TN RDI30 RDI30 TP TP + FN TN TN + FP Positive accuracy TP TP + FP 49

62 Negative accuracy 5.2 TN TN + FN

63 5.4 RDI30 30 Skewness Skewness Skewness 51

64 5.8 Skewness Skewness 52

65 V feature 5.10 V feature V feature 5.11 V feature 53

66 V feature 5.12 V feature V feature 5.13 V feature 54

67 V feature 5.14 Skewness V feature V feature 5.15 Skewness V feature 55

68 V feature 5.16 V feature V feature 5.17 V feature 56

69

70

71 5.22 Skewness Skewness RDI

72 RDI Histogram % RDI % RDI % RDI 60

73 5.24 RDI 30~ RDI 20~ data base RDI RDI RDI RDI 0~5 55 RDI 0~ ~ ~ RDI

74 ~ ~9 0~ %

75 ~

76 0~ Skewness Skewness 5.34 Skewness 5.35 Skewness 64

77 Skewness

78 ~ ~ V feature V feature 66

79 5.40 V feature 5.41 V feature V feature 0~40 V feature V feature ~

80 RDI 17.4~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~

81 SpO 2 RDI Vázquezz 30 OSA 87.3 RDI RDI OSA RDI 15 OSA OSA OSA 30 OSA RDI

82 Atiya AF, El-Shoura SM, Shaheen SI and El-Sherif MS, 1999, A comparison between neural-network forecasting techniques-case study: river flow forecasting, IEEE Transactions on Neural Networks, vol. 10(2), pp Atul Malhotra and David P White, 2002, Obstructive sleep apnoea, Lancet, vol. 360, pp Behbehani K, Yen FC, 1995, Automatic control of airway pressure for treatment of obstructive sleep panea, IEEE Transactions on Biomedical Engineering, vol. 42, NO. 10, pp Behbehani K, Yen FC, Axe J, Burk J, and Lucas E, 1993, Adaptive positive airway pressure therapy for obstructive sleep apnea, IEEE Transactions on Biomedical Engineering, pp Cao L, Tay FEH, 2001, Financial forecasting using support vector machines, Neural Comput & Applic, vol. 10, pp Charylonink W and Chen MS, 2000, Very short-term load forecasting using artificial neural networks, IEEE Transactions on Power Systems, vol. 15(1), pp

83 Haykin S, 1999, Neural Networks: A Comprehensive foundation, 2 nd ed. Englewood Cliffs, NJ: Pretice-Hall. Juan-Carlos V, Willis HT, W Ward F, Akira M, Rollin B, Eric H, William AW, John ER, 1999, Automated analysis of digital oximetry in the diagnosis of obstructive sleep apnoea, Thrax 2000, vol. 55, pp Lavie P, Here P, Hoffstein V, 2000, Obstructive sleep apnoea syndrome as a risk factor for hypertension, BMJ, vol. 320, pp Leo, B., 1996, Bagging Predictors, Machine Learning, 24, pp Marcelo CM and Álvaro V, 2000, A hybrid linear-neural model for time series forecasting, IEEE Trans. Neural Networks, vol. 11(6), pp Montserrat JM, Rarre R, Navajas D, 2001, Automatic continuous positive airway pressure devices for the treatment of sleep apnea hypopnea syndrome, Sleep Medicine, vol. 2, pp Paola Lanfranchi and Virend K Somers, 2001, Obstructive sleep apnea and vascular disease, Respir Res, vol. 2, pp Pattie DC and Snyder J, 1996, Using a neural network to forecast visitor behavior, Annals of Tourism Research, vol. 23(1),

84 Penzel T, 2000, The apnea-ecg database, in Computers in Cardiology, Piscataway, NJ:IEEE, vol. 27, pp Ranta SOV, Hynynen M and Räsänen J, 2002, Application of artificial neural networks as an indicator of awareness with recall during general anaesthesia, Journal of Clinical Monitoring and Computing, vol. 17, pp Robert C, Karasinski P, Arreto CD and Gaudy JF, 2002, Monitoring anesthesia using neural networks: a survey, Journal of Clinical Monitoring and Computing, vol. 17, pp Rosenthal L, Gerhardstein R, Lumley A, Guido P, Day R, Syron ML, Roth T, 2000, CPAP therapy in patients with mild OSA:implementation and treatment outcome, Sleep Medicine, vol. 1, pp Rossiev DA, Golovenkin SE, Shulman VA and Matjushin GV, 1995, Neural networks for forecasting of myocardial infarction complications, 2 nd International Symposium on Neuroinformatics and Neurocomputers, pp. Teran-Santos J, Jimenez-Gomez A, Cordero-Guevara J, 1999, The association between sleep apnoea and the risk of traffic accidents, N Engl J Med, vol. 340, pp

85 Ulysses JM, Jacek D, Sateesh V, Azmi D, M. Jeffery M, Ali E, and Brydon J. B. G, 2003, Prediction of the Apnea-Hypopnea Index From Overnight Pulse Oximetry, Chest, vol. 124, pp Wilson H and Recknagel F, 2001, Towards a generic artificial neural network model for dynamic predictions of algal abundance in freshwater lakes, Ecological Modelling, vol. 146, pp Yen FC, Khosrow MS, John R, Burk MD, 1996, Long trem performance evaluation of an automatic airway positive pressure device, IEEE Engineering in Medicine and Biomedical Society, pp Yoav, F., and Robert, E. S., 1997, A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting, Journal of Computer and System Sciences, 55, pp Young T, Palta M, Dempsey J, Skatrud J, Weber S, Badr S, 1993, The occurrence of sleep-disordered breathing among middle-aged adults, N Engl J Med, vol. 328, pp Young T, Peppard P, Palta M, Hla KM, Finn L, Margan B, and Skatrud J, 1997, Population-based study of sleep-disordered breathing as a risk factor for hypertension, Arch. Intern. Med., vol. 157, pp , 2004,,. 73

86 , 2002,,., 2004,,,. 74