應用K-nearest Neighbor分類法診斷異常肺音

Using K-nearest Neighbor Classification to Diagnose Abnormal Lung Sounds

陳奕安(Yi-An Chen)

指導教授 ： 譚旦旭

共同指導教授 ： 黃文增(Wen-Tzeng Huang)

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2014年)

摘要

異常肺音中的爆裂音(Crackles)、乾囉音(Rhonchi)及喘鳴音(Wheeze)所關聯的人口總數約占全球人口30%，由於目前的聽診儀存在許多缺點，如(1)聽診結果受醫師主觀經驗控制，使得診斷難有一致標準、(2)聽診結果易受測量環境噪音影響及(3)無法儲存心肺音，不易追蹤病情。因此本研究應用分類技術，針對爆裂音、乾囉音及喘鳴音建構一套肺音疾病聽診輔助系統。我們使用梅爾頻率倒頻譜係數(Mel-frequency Cepstral Coefficients, MFCC)擷取正常肺音與異常肺音(Crackle, Wheeze及Rhonchi)的特徵參數，再結合K-平均值演算法(K-means Algorithm)對訊號特徵參數進行聚類，以降低資料量及計算量。最後，透過前K位最鄰近法(K-nearest Neighbor)，對異常肺音訊號進行分類。針對居家照顧之應用，當異常肺音區段達到30%，會發出建議用戶就醫的訊息。
另外，本研究開發彎曲感測器來偵測受測者呼吸狀態，此呼吸偵測器具有高準確度的偵測能力，量測結果與實際值誤差約6.8%。彎曲感測器的訊號可透過無線藍芽傳輸到電腦計算其呼吸週期（次卅分鐘）以達到即時偵測的功效。當使用者呼吸週期發生異常時，系統會發出警訊。本研究的肺音異常診斷系統與無線呼吸偵測系統，可以提供居家自我照護的應用，具備實用潛力。

關鍵字

數位式聽診器 ； 肺音訊號 ； K-平均值演算法 ； 前K位最鄰近法 ； 梅爾頻率倒頻譜係數 ； 呼吸週期

並列摘要

As reported, 30% people worldwide have the symptom of abnormal lung sounds including crackles, rhonchi, and wheeze. Currently, the traditional stethoscope is the most popular tool used by doctor for diagnosing abnormal lung sounds. However, many problems remain when using the traditional stethoscope, which include adverse effect due to environmental noise, unable to store the lung sound for follow-up tracking, and doctor’s subjective diagnosis experiences. To overcome the above mentioned problems, a digital stethoscope is presented in this study to help doctor in diagnosis of abnormal lung sounds.
In this proposed digital system, the Mel-frequency Cepstral Coefficients (MFCC) is employed to extract the feature of lung sound and then the k-means Algorithm is applied for feature clustering in order to reduce the amount of data for computation. Finally, the K-nearest Neighbor is utilized to classify the lung sounds. The proposed system can also be used for homecare; as the percentage of abnormal lung sounds frames is over 30% of a whole test signal, the system will automatically warn the user to visit doctor for diagnosis.
Additionally, we employ bend sensors together with amplification circuit, Bluetooth, and microcontroller to implement a respiration detector. The respiratory signal extracted by bend sensors can be transmitted to the computer via Bluetooth to calculate the respiratory cycle for real-time detection. As abnormal status happens the detector will warn the user automatically. Experimental result indicates that the error of respiratory cycle between the measured value and real value is only 6.8%, which illustrates the potential of our proposed detector for homecare application.

並列關鍵字

Stethoscope ； Lung Sound ； K-means Algorithm ； MFCC ； K-nearest Neighbor

參考文獻

[35] 鄧乃睿，心音訊號有效區間之擷取及其於數位聽診系統之應用，碩士論文，國立台北科技大學，台北，2012。

[18] 蔡秉翔，建立於隨身聽的數位式電子聽診器，國立中央大學，2007。

[17] 余勝智，肺音感測系統之設計與實現，國立臺灣師範大學機電科技學系碩士論文，2008。

[45] 徐慶豐，雙模式藍芽數位聽診儀之介面設計，碩士論文，國立台北科技大學，台北，2013。

[20] J. F. Murray and J. A. Nadel, “Textbook of Respiratory Medicine,” Philadelphia: W. B. Saunders, 1994.