以HHT為基礎之肺音分析與哮喘音辨識研究

HHT-Based Lung Sound Analysis and Wheezing Recognition

楊佳穎(Chia-Ying Yang)

指導教授 ： 李仁貴

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2008年)

https://doi.org/10.6841/NTUT.2008.00200

摘要

哮喘音(Wheeze)為氣喘診斷的重要訊息之一，一般醫師常用聽診法聽取肺音資訊做初步診斷，但因無法記錄肺音訊號與相關資訊，且須經過醫師這類的專業人員來判讀，無法廣泛應用於居家或做長時的監控，因此發展一個可應用在小型裝置上的氣喘音辨識演算法有其必要性。
基於肺音非穩態、非線性的特性，本論文利用希爾伯特-黃轉換(Hilbert-Huang Transformation, HHT) 訊號分析法來做處理。原始肺音資料透過經驗模態分解法，將信號展開成數個本質模態函數（Intrinsic Mode Function, IMF），再利用希伯特轉換求得IMF的瞬時頻率及振幅，由時頻圖可看出肺音信號瞬時變化的訊息與特性。
本系統取樣頻率為 5000 Hz，分類法採用簡單的條件判斷法，取IMF1、IMF2對應之瞬時頻率，經由移動平均法(Moving Average)量化瞬時頻率，透過閥值區分出哮喘音與正常聲音的差別。
為將演算法實際應用在病人上，透過一市售聽診器結合錄音筆作記錄，實際量測氣喘病人之肺音，每記錄檔約1分鐘，經由低通濾波器濾除雜訊與重新取樣等前處理後，套用本演算法可達高正確率。
本論文提出之方法運算簡單，可保留聲音時、頻域的特徵方便觀察，且符合美國胸腔協會(American Thoracic Society, ATS)定義之標準，正確率高達93.33%，未來期可應用在小型嵌入式系統作居家照護與長時監控上。

關鍵字

肺音 ； 哮鳴音 ； HHT ； 訊號處理 ； 氣喘

並列摘要

Auscultation of pulmonary sounds provides valuable clinical information but has been regarded as a tool with low diagnostic value due to the inherent subjectivity in the evaluation of these sounds. And it does not allow a permanent record of data, long-term monitoring of pulmonary sounds in follow-up studies is not possible. For the reasons, the development of algorithms which are used to auto-calculus and self-recognized measures is essential.
As a result of the astaticism and nonlinear characteristics of lung sounds signals, a new approach in the analysis of the nonlinear characteristics of wheezes based on Hilbert-Huang Transformation (HHT) was presented in this study.
We use the HHT algorithm to estimate the relations between frequency, time, and amplitude about the basic lung sounds signals. The HHT algorithm is used to capable of acquiring, parameterize and subsequently classifying lung sounds into anomalistic and normal sounds with an aim to evaluate them objectively.
The sampling rate of 5000 Hz is chosen in this system. We could recognize the wheeze from normal condition by instantaneous frequency(IF) of IMF1 and IMF2（Intrinsic Mode Function, IMF）using Moving Average quantification .
The lung sounds of asthmatic patient are converted into analog signals via a commercial stethoscope and can be stored in a commercial recorder. After the pretreatment by filtered signal and sampling processing, we used the HHT algorithm to recognize the part of aberrant voice.
In this paper, we bring a not only novel but easier way to keep the transfer function in time-frequency domain of the lung sounds. Experiments show that the method has high accuracy. Furthermore, the proposed method is fully automated without any additional need for adjusting the method to respiratory subjects.
In the future we can make a home-care service embedded system and monitoring patients as long as possible.

並列關鍵字

lung sound ； HHT ； signal processing ； wheeze ； asthma

參考文獻

[20] 陳冠宏，多種肺音自動辨識之研究，碩士論文，國立台灣大學生物產業機電工程學研究所，台北市，2005。

[1] The Global Initiative for Asthma (GINA), http://www.ginasthma.com/.

[4] K. A. Kim, J. H. Lee, T. S. Lee, E. J. Cha, “Peak expiratory flow meter capable of spirometric test for asthma,” Proceedings of IEEE Asian-Pacific Conference on Engineering in Medicine and Biology Society, pp. 254-255, 2003.

[5] Y. Shabtai Musih, B. G. James, and G. Noam, “Spectral content of forced expiratory wheezes during air, He, and SF6 breathing in normal humans,” Journal of Applied Physiology, vol. 72, no. 2, pp. 629-635, 1992.

[7] A. Homs Corbera, R. Jane, J. A. Fiz, and J. Morera, “Algorithm for time-frequency detection and analysis of wheezes,” Proceedings of the 22nd IEEE Annual International Conference on Engineering in Medicine and Biology Society, vol. 4, pp. 2977-2980, 2000.