跳到主要內容

臺灣博碩士論文加值系統

(44.221.73.157) 您好!臺灣時間:2024/06/17 21:52
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:謝譯瑭
研究生(外文):Yi-Tang Hsieh
論文名稱:基於心音圖時頻域特徵的心臟病自動檢測模型
論文名稱(外文):An automatic detection model for cardiovascular disease based on time-frequency domain features of PCG signals
指導教授:郭美惠郭美惠引用關係
指導教授(外文):Guo,Mei-Hui
學位類別:碩士
校院名稱:國立中山大學
系所名稱:應用數學系研究所
學門:數學及統計學門
學類:數學學類
論文種類:學術論文
論文出版年:2023
畢業學年度:111
語文別:英文
論文頁數:84
中文關鍵詞:深度學習機器學習心音圖時頻分析極限梯度提升
外文關鍵詞:Deep LearningMachine LearningPhonocardiogramTime-Frequency AnalysisXGboost
相關次數:
  • 被引用被引用:0
  • 點閱點閱:32
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
心血管疾病是現今常見且致命的疾病之一,因此及早發現並治療對於提升患者的存活率至關重要。聽診器是一種常見的診斷工具,透過聽診器收集的心音轉換成心音圖(PCG)。然而,即使是經驗豐富的醫生,也很容易出現誤診的情況。因此,在過去的幾十年中,透過分析心音圖並結合機器學習的方法來建立分類模型一直是熱門的研究主題。儘管一些關於心音分類的研究表現出相當優異的結果,但這些研究存在一些缺失。其中一個缺失是大多數研究僅考慮乾淨的資料,在完全理想的情況下進行研究,這與臨床實際情況不符。因此,本研究提出了一種新的方法,使用時頻域分析提取多重特徵,再結合卷積遞迴神經網絡模型和機器學習模型,以提高心音辨識的準確性和醫學實用性。

我們將所提出的方法應用在所蒐集的838筆臨床醫學資料中,並使用深度學習結合XGboost的模型得到的分類結果,在測試集上達到了85.26%的準確率(敏感度:79.22%,特異度:87.93%)。相較於醫師診斷的結果(81.04% 的準確率。敏感度:68.98%,特異度:98.65%),顯示出我們提出的方法具有更高的準確度和敏感度。最後,為了驗證模型的穩健性和可靠性,我們將所提出的方法應用於著名的2016 PhysioNet/CinC Challenge公開資料集,其中深度學習結合XGboost得到了95%的準確率(敏感度:91.81%,特異度:98.88%)。
Cardiovascular disease is one of the leading causes of death today. Therefore, early detection and treatment are essential to improve patient survival rates. The stethoscope is a common diagnostic tool that converts heart sounds collected through the stethoscope into a phonocardiogram (PCG). However, even experienced doctors can sometimes make misdiagnoses. Therefore, in the past few decades, it has been a popular research topic to develop automatic heart sound detection models by analyzing phonocardiograms and combining machine learning methods. Although some studies have shown good results, yet there are still shortcomings, such as most studies only consider clean data and conduct research under ideal conditions, which don''t correspond to the clinical situation. In this study, we propose a new method that uses time-frequency domain analysis to extract multiple features and then combines a stacked recurrent neural network model and machine learning methods to improve the accuracy and practicality of heart sound recognition.

We apply the proposed methods to 838 clinical data; the results show that a deep learning model combined with XGboost obtains the best detection results, achieving an accuracy of 85.26% (sensitivity: 79.22%, specificity: 87.93%) on the test set. Compared with the results of experienced cardiologists, achieving an accuracy of 81.04% (sensitivity: 68.98%, specificity: 98.65%), our proposed method has higher accuracy and sensitivity. We apply the proposed approach to the famous 2016 PhysioNet/CinC Challenge public dataset to further validate the model''s reliability. Deep learning combined with XGboost achieves an accuracy rate of 95% (sensitivity: 91.81%, specificity: 98.88%).
論文審定書 i
誌謝 ii
摘要 iii
Abstract iv
1 Introduction 1
1.1 Heart structure and cardiac cycle . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Introduction of heart murmur . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Data Description 8
2.1 Clinical data from E-DA hospital . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Heart disease detection using audios at location 1~5 . . . . . . . 10
2.1.2 Heart disease detection using audios at specific location . . . . . 13
2.2 Open data PhysioNet/CinC challenge 2016 . . . . . . . . . . . . . . . . 15
3 Methodology 16
3.1 Signal processing procedure . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.1 Butterworth band-pass filter . . . . . . . . . . . . . . . . . . . . 17
3.1.2 Downsampling and Normalization . . . . . . . . . . . . . . . . . 19
3.2 Segmentation and decision guidelines . . . . . . . . . . . . . . . . . . . 22
3.3 Feature extraction methods . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3.1 Mel Frequency Cepstrum Coefficient . . . . . . . . . . . . . . . 24
3.3.2 Short time Fourier transform . . . . . . . . . . . . . . . . . . . . 32
3.3.3 Deep Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.4 Constant-Q transform . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Network models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4.1 Residual Networks (ResNet) . . . . . . . . . . . . . . . . . . . . 47
3.4.2 Self-attention mechanism . . . . . . . . . . . . . . . . . . . . . . 48
3.4.3 Long short-term memory model . . . . . . . . . . . . . . . . . . 50
3.4.4 Weights of each feature . . . . . . . . . . . . . . . . . . . . . . . 53
4 Empirical study 55
4.1 Heart disease detection using audios at location 1~5 . . . . . . . . . . . . 57
4.1.1 Individual M, S, D, C +ResRNN . . . . . . . . . . . . . . . . . . 57
4.1.2 A + ResRNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.3 A + ResRNN + Machine Learning . . . . . . . . . . . . . . . . . 58
4.1.4 Visualization of attention maps . . . . . . . . . . . . . . . . . . 60
4.2 Heart disease detection using audios at specific location . . . . . . . . . . 62
4.2.1 A + ResRNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2.2 A + ResRNN + Machine Learning . . . . . . . . . . . . . . . . . 63
4.3 Verification on PhysioNet/CinC challenge 2016 . . . . . . . . . . . . . . 65
4.4 Analytic platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5 Conclusion and discussion 68
6 Future work 69
A Appendix 70
References 73
Abdul, Z. K., & Al-Talabani, A. K. (2022). Mel frequency cepstral coefficient and its
applications: A review. IEEE Access, 10, 122136–122158. https:// doi. org/ 10.
1109/ACCESS.2022.3223444
Andén, J., & Mallat, S. (2014). Deep scattering spectrum. IEEE Transactions on Signal
Processing, 62(16), 4114–4128. https://doi.org/10.1109/TSP.2014.2326991
Balduzzi, D., Frean, M., Leary, L., Lewis, J., Ma, K. W.-D., & McWilliams, B. (2018). The
shattered gradients problem: If resnets are the answer, then what is the question?
Britannica, T. E. o. E. (2023). Heart. encyclopedia britannica. Retrieved June 7, 2023,
from https://www.britannica.com/science/heart
Brown, J. C. (1991). Calculation of a constant q spectral transform. Journal of the Acous tical Society of America, 89, 425–434.
Chengyu Liu, e. a., David Springer. (2016). Classification of heart sound recordings: The
physionet/computing in cardiology challenge 2016.
Diez, M., Penagarikano, M., Bordel, G., Varona, A., & Rodriguez-Fuentes, L. J. (2014). On
the complementarity of short-time fourier analysis windows of different lengths for
improved language recognition. https://doi.org/10.21437/Interspeech.2014-608
He, K., Zhang, X., Ren, S., & Sun, J. (2015). Deep residual learning for image recognition.
CoRR, abs/1512.03385. http://arxiv.org/abs/1512.03385
Huang, P.-K., Yang, M.-C., Wang, Z.-X., Huang, Y.-J., Lin, W.-C., Pan, C.-L., & Guo,
M.-H. (n.d.). Augmented detection of septal defects using advanced optical coher ence tomography network-processed phonocardiogram. Frontiers in Cardiovas cular Medicine, 9. https://doi.org/10.3389/fcvm.2022.1041082
Janse, P., Magre, S., Kurzekar, P., & Deshmukh, R. (2014). A comparative study between
mfcc and dwt feature extraction technique.
Jiang, W., Wu, X., Wang, Y., Chen, B., Feng, W., & Jin, Y. (2021). Time–frequency analysis-based blind modulation classification for multiple-antenna systems. Sen sors, 21, 231. https://doi.org/10.3390/s21010231
Jiang, Z., Choi, S., & Wang, H. (2007). A new approach on heart murmurs classification
with svm technique, 240–244. https://doi.org/10.1109/ISITC.2007.40
Jung, S.-Y., Liao, C.-H., Wu, Y.-S., Yuan, S.-M., & Sun, C.-T. (2021). Efficiently clas sifying lung sounds through depthwise separable cnn models with fused stft and
mfcc features. Diagnostics, 11(4). https://doi.org/10.3390/diagnostics11040732
Kusko, M. C., & Maselli, K. (2015). Introduction to cardiac auscultation. In A. J. Taylor
(Ed.), Learning cardiac auscultation: From essentials to expert clinical interpreta tion (pp. 3–14). Springer London. https://doi.org/10.1007/978-1-4471-6738-9_1
LeBlond RF, B. D., DeGowin RL. (2008). Degowin’s diagnostic examination, 9th edition.
MOHW. (2022). Cardiovascular diseases news. Retrieved July 8, 2023, from https : / /
www.mohw.gov.tw/cp-4634-52471-1.html
Muda, L., Begam, M., & Elamvazuthi, I. (2010). Voice recognition algorithms using mel
frequency cepstral coefficient (MFCC) and dynamic time warping (DTW) tech niques. CoRR, abs/1003.4083. http://arxiv.org/abs/1003.4083
Navarro, L., Courbebaisse, G., & Jourlin, M. (2014). Chapter two - logarithmic wavelets
(P. W. Hawkes, Ed.). 183, 41–98. https://doi.org/https://doi.org/10.1016/B978-0-
12-800265-0.00002-3
Richard E. Klabunde, P. (2023). Cardiovascular physiology concepts. Retrieved January
26, 2023, from https://cvphysiology.com/heart-disease/hd008
S.Butterworth et al. (1930). On the theory of filter amplifiers. Wireless Engineer, 7(6),
536–541.
Singh, P., Saha, G., & Sahidullah, M. (2021). Non-linear frequency warping using constant q transformation for speech emotion recognition, 1–6. https://doi.org/10.1109/
ICCCI50826.2021.9402569
Strunic, S., Rios-Gutierrez, F., Alba-Flores, R., Nordehn, G., & Burns, S. (2007). De tection and classification of cardiac murmurs using segmentation techniques and
artificial neural networks, 397–404. https://doi.org/10.1109/CIDM.2007.368902
Tiwari, S., Jain, A., Sharma, A. K., & Mohamad Almustafa, K. (2021). Phonocardiogram
signal based multi-class cardiac diagnostic decision support system. IEEE Access,
9, 110710–110722. https://doi.org/10.1109/ACCESS.2021.3103316
Todisco, M., Delgado, H., & Evans, N. (2017). Constant q cepstral coefficients: A spoofing
countermeasure for automatic speaker verification. Computer Speech Language,
45, 516–535. https://doi.org/https://doi.org/10.1016/j.csl.2017.01.001
Vyas, S., Patil, M., & Birajdar, G. (2021). Classification of heart sound signals using time frequency image texture features. https://doi.org/10.1002/9781119818717.ch5
WHO. (2023). Cardiovascular diseases. Retrieved July 8, 2023, from https://www.who.
int/health-topics/cardiovascular-diseases#tab=tab_1
Zhang, A., Lipton, Z. C., Li, M., & Smola, A. J. (2021). Dive into deep learning. CoRR,
abs/2106.11342. https://arxiv.org/abs/2106.11342
電子全文 電子全文(網際網路公開日期:20280814)
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊