- 學位論文
中醫聞診自動分析-虛症患者語音之分類辨識
A Study of automatic analysis for Listening Examination of Chinese Medicine - Classification and Recognition In vacuity patient's voice
摘要
本論文主要研究目的，是建立可攜式聞診語音系統，並針對非虛、氣虛與陰虛證型受測者之語音訊號進行分類與辨識。 傳統中醫以「望、聞、問、切」四診合參，進行臨床辨證診斷。聞診方面，主要以耳聽聲音與鼻嗅氣味為診斷方法。由於診斷過程往往須憑藉中醫師臨床經驗進行辨症，導致結果不夠客觀且缺乏說服力。目前聞診相關研究，主要仍以分析受測者語音訊號為主。本研究與林口長庚醫院中醫診斷研究室合作，收錄50名非虛、30名氣虛、29名陰虛及12名心臟衰竭患者語音訊號/a/。擷取受測者語音訊號之單一週期平均訊號，利用雙向關聯式記憶網路進行語音訊號波形分類，最後再對波形作定量化參數分析，找出語音訊號差異性。在聞診語音系統設計方面，以個人數位助理為架構，透過內建麥克風擷取受測者語音，方便中醫師於臨床應用。 分析結果，非虛與氣虛識別方面，平均正確率為82%，氣虛患者語音波形在副波強度部份明顯較非虛者來得弱。在非虛與陰虛識別方面，平均正確率為80%，並發現兩者語音波形之間，在頻譜能量方面有較顯著差異。最後將12名心臟衰竭患者之語音訊號進行分類測試，結果有75% 與中醫師臨床辨證的結果一致。
並列摘要
The aim of this study is to develop a portable listening and analysis system for non-vacuity, qi-vacuity and yin-vacuity patients’ voices. In listening examination, the method used in diagnosis is divided into listening and smelling. Up to now, the major study in listening examination still emphasizes on analyzing patient’s voices. We collected 121 patient’s voice samples (50 was non-vacuity patients, 30 was qi-vacuity patients, 29 was yin-vacuity patients, and 12 was CHF (congestive heart failure) patients). We extracted single period wave of patient’s voice and classified them with Bi-directional Associative Memory (BAM) network. Listening system architecture is based on personal digital assistant (PDA). It will record the patient’s voice through a microphone embedded in PDA. According to analysis result, we came up with the mean accuracy of 82% in recognizing non-vacuity and qi-vacuity. The voice waveform of qi-vacuity patients is weaker than non-vacuity in minor-wave intensity. We have mean accuracy 80% in recognizing non-vacuity and yin-vacuity. We also find out that the power spectrum has clear difference between the two voice waveform. At last, we held a test to classify twelve CHF patient’s voice signals and came up with a 75% accuracy comparing to physician’s diagnosis.