臺灣博碩士論文加值系統

近年來已經有許多評估麻醉患者的意識程度的儀器已經被開發出來，其中最常用的商業患者監測儀器產品是Philips IntelliVue Patient monitor。但是目前評估患者麻醉深度的方法都是透過複雜的數值計算得到的指標，本研究採用近年來快速崛起的深度學習中的convolution neural network (CNN)方法來判斷麻醉患者的麻醉程度。
本研究中是用Philips MP60儀器收集臨床患者的electroencephalogram (EEG)原始數據，EEG數據會採用STFT和CWT演算法將原始號轉換成時頻影像。影像會根據MP60給出的Bi-Spectral (BIS)值和signal quality indicator (SQI)作為分類腦波影像的標準，類別會被分為Anesthetic Light、Anesthetic Ok、Anesthetic Deep 和 Noise。
在本研究中執行了兩項實驗，第一項實驗是對使用相同麻醉藥的13名患者EEG訊號每5秒以及2分鐘經過CWT算法轉換成影像，用影像對CNN中的Alexnet模型進行訓練並比較結果，結果長時間影像訓練的模型準確率較高。第二項實驗針對特定的手術與提升患者的樣本數，以不同的方式轉換原始訊號，由55名患者以2分鐘的EEG訊號透過CWT和STFT算法轉換成影像，CNN模型分別會對兩種不同的時頻影像進行訓練，最後測試不同算法影像的模型整體準確率與4個類別的準確率。第二項實驗會在從 55 名患者隨機選出十名患者作為真實情況下的模型實測。最後再用交叉驗證的方式來評估模型的泛化性。整個研究結果顯示長期的EEG訊號影像有助於CNN特徵提取的優勢，另外CWT影像模型準確率略勝於STFT影像模型。

In recent years, many instruments have been developed to assess the depth of anesthesia of anesthetized patients. The most commonly used monitoring instrument for monitoring anesthesia patients is the Philips IntelliVue Patient monitor. However, the current methods for assessing the depth of anesthesia are based on complex numerical calculations. This study applied the convolutional neural network (CNN) to assess the depth of anesthesia.
In this study, clinical patients' electroencephalogram (EEG) raw signal was collected by Philips MP60 instrument. The EEG signal was converted to a time-frequency image using the STFT and CWT algorithms. The image is based on Bi-Spectral (BIS) values and signal quality indicators (SQI) as criteria for classifying images. The categories will be classified into Anesthesia Light, Anesthesia Ok, Anesthesia Deep and Noise.
Two experiments were performed in this study. The first experiment was 13 patients using the same anesthetic. Patient EEG signal is converted to image by CWT algorithm every 5 seconds and 2 minutes. Use images to train the Alexnet model and compare test results. As a result, the accuracy of the model for long-term image training is higher. The second experiment targeted specific surgery and increased patient data then converted long-term EEG signals into images. 55 patients’ EEG signal was converted to image by CWT and STFT algorithms. Train the model with two different images and compare the overall accuracy and accuracy of the four categories. Ten patients were randomly selected from the data of 55 patients as the actual implementation results of the test model. Finally, cross-validation is used to evaluate the generalization of the model. The entire research results show that the long-term EEG signal image contributes to the advantages of CNN feature extraction. In addition, the accuracy of the CWT image model is slightly better than the STFT image model.

A Thesis ii
摘 要 iii
ABSTRACT iv
Contents v
List of Tables vi
List of Figures vii
Abbreviation viii
Chapter 1 Introduction 1
1.1 Background 1
1.2 Motivation 2
1.3 Purpose 3
1.4 Brief summary of chapters 4
Chapter 2 General Anesthesia 5
2.1 Anesthesia process 5
2.2 Electroencephalography (EEG) 6
2.3 CNN application 7
2.4 Confusion matrix 8
Chapter 3 Algorithm 10
3.1 Continuous wavelet transform (CWT) 10
3.2 Short-time Fourier transform (STFT) 11
3.3 Convolution neural network (CNN) 12
Chapter 4 Experiment Condition and Method 14
4.1 Data sources 14
4.2 Data process 16
4.3 Super computer operation 21
4.4 Cross-validation 22
Chapter 5 Result 24
5.1 CWT image model test results 24
5.2 The first experiment result 26
5.3 The second experiment result 28
5.3 Randomly select patients to test model 31
5.4 Cross-validation result 34
Chapter 6 Conclusion and Future Work 41
References 43

1. G. Kotsovolis and G. Komninos, “Awareness during anesthesia: how sure can we be that the patient is sleeping indeed?” Hippokratia, 2009, 13: 83–9.
2. A. Petsiti, V. Tassoudis, G. Vretzakit, D. Zacharoulis, K. Tepetes, G. Ganeli, et al. “Depth of anesthesia as a risk factor for perioperative morbidity,” Anesthesiol Res Pract, 2015,2015:829151.
3. J. C. Drummond, C. A. Brann, D. E. Perkins, and D. E. Wolfe, “A comparison of median frequency, spectral edge frequency, a frequency band power ratio, total power, and dominance shift in the determination of depth of anesthesia,” Acta Anaesthesiologica Scandinavica, 1991, 35(8), 693-699.
4. C. D. Kent, and K. B. Domino, “Depth of anesthesia,” Curr Opin Anaesthesiol, 2009, 22(6), 782-787.
5. M. Jeanne, R. Logier, J. D. Jonckheere and B. Tavernier, “Heart rate variability during total intravenous anesthesia: effects of nociception and analgesia,” Autonomic Neuroscience, 2009, 147(1), 91-96.
6. C. J. Pomfrett, S. Dolling, N. R. Anders, D. G. Glover, A. Bryan, and B. J. Pollard, “Delta sleep-inducing peptide alters bispectral index, the electroencephalogram and heart rate variability when used as an adjunct to isoflurane anesthesia,” European Journal of Anaesthesiology (EJA), 2009, 26(2), 128-134.
7. J. Bruhn, H. R#westeur055#pcke, and A. Hoeft, “Approximate entropy as an electroencephalographic measure of anesthetic drug effect during anesthesia,” Anaesthesiology 2000, 92, pp.715–726.
8. H. Vierti#westeur055#-Oja, V. Maja, M. S#westeur037#rkel#westeur037#, P. Talja, N. Tenkanen, H. Tolvanen-Laakso, M. Paloheimo, A. Vakkuri, A. Yli-Hankala, and P. Meril#westeur037#inen, “Description of the Entropy™ algorithm as applied in the Datex-Ohmeda S/5™ entropy module Acta Anaesthesiol,” Scand, 2004, 48, pp.154–61.
9. J.C. Sigl, and N.G. Chamoun, “An introduction to bispectral analysis for the electroencephalogram,” Journal of Clinical Monitoring, vol. 10, no. 6, 1994, pp.392–404.
10. L. C. Jameson and T. B. Sloan, “Using EEG to monitor anesthesia drug effects during surgery,” Journal of clinical monitoring and computing, 2006, 20(6), 445-472.
11. S.A. Taywade and R.D. Raut, “A review: EEG signal analysis with different methodologies,” in Proceedings of the National Conference on Innovative Paradigms in Engineering and Technology (NCIPET ’12), 2014, pp.29–31.
12. K. Kuizenga, JM. Wierda, and CJ. Kalkman, “Biphasic EEG changes in relation to loss of consciousness during induction with thiopental, propofol, etomidate, midazolam or sevoflurane,” Br J Anaesth, 2001, 86, pp.354–360.
13. B. Boashash, M. Mesbah, and P. Golditz, “Time-Frequency Detection of EEG Abnormalities,” Amsterdam, The Netherlands: Elsevier, 2003, pp.663–669, ch15.
14. G. Hinton and R. Salakhutdinov, “Reducing the dimensionality of data with neural networks,” Science Magazine, 2006, vol. 313, pp. 504-507.
15. L. Wei, Y. Lin, J. Wang, and Y. Ma, “Time-frequency convolutional neural network for automatic sleep stage classification based on single-channel EEG,” in 2017 IEEE 29th International Conference on Tools with Artificial Intelligence (ICTAI), 2017, pp. 88–95.
16. H. Adeli, “Deep convolutional neural network for the automated detection and diagnosis of seizure using EEG signals,” Comput. Biol. Med., Sep. 2017, vol. 100, pp. 270–278.
17. S. Tripathi, S. Acharya, R.D. Sharma, S. Mittal, and S. Bhattacharya, “Using deep and convolutional neural networks for accurate emotion classification on DEAP dataset,” in Proc. IAAI, 2017, pp.4746–4752.
18. Z. Tang, C. Li, and S. Sun, “Single-trial EEG classification of motorimagery using deep convolutional neural networks,” Optik Int JLight Electron Opt, 2017, 130:11–18.