Logo image
呼吸生理訊號之分析在醫療上的應用
Thesis

呼吸生理訊號之分析在醫療上的應用

王遵偉
Masters, National Tsing Hua University
1993

Abstract

特徵抽取;類神經網路;自迴歸模型;潮氣容量;誤差逆傳遞 Feature extraction Neural networks Autoregressive model Ti l volume Error backpropagation
本篇論文提供了一些呼吸生理訊號的分析方法,嘗試建立呼吸生理訊號的特徵,以應用於呼吸道疾病的臨床診斷與監護。不同於心電圖訊號(ECG),呼吸波形沒有一定的軌跡可循。 即使是同一個人, 在不同時間所取得的呼吸訊號均不相同。 因此, 要分辨隱含不同呼吸道疾病的呼吸波形,就必須從尋找適當的呼吸特徵著手。本篇論文分別從時域 (timedomain) ,頻域 (frequency domain),以及系統模擬 (systemmodeling) 來建立呼吸特徵。利用不同疾病下呼吸特徵的差異性,來達到呼吸道疾病鑑定的目的。此篇論文研究的是一般性且整體化的問題解決方案(general solutio n)。從呼吸感測器的選擇, 數位化(A/D)及放大電路的選用設計,呼吸監護訊號的轉換以達成呼吸資料的汲取,再使用本文提出的方法來作特徵抽取,進而利用類神經網路(Artificial NeuralNetworks; ANN) 來作特徵訓練分類,最後根據分類的結果提出一些呼吸醫療監護上的應用以及呼吸訊號處理的結論。在傳統呼吸系統功能的臨床診斷上,醫生往往必須藉助一些昂貴而且需要病人配合的生理檢驗儀器,來量測病人的呼吸生理資料,以作為診斷的輔助參考。而另一方面,行之有年的呼吸監護系統,則是利用一些呼吸感測裝置,如溫度,阻抗等感測器,將病人的呼吸行為,轉化成呼吸波形並顯示在螢幕上,以提供醫護人員監護的目的。本論文的目的,則是希望將呼吸監護系統加上波形分析的功能,使其能夠具備呼吸系統疾病初步診斷的能力,作為醫師輔助診斷的參考。 同時, 此系統也能偵測不正常的呼吸波形以通知醫護人員,成為智慧型的監護系統。這項研究的嘗試不但具有前瞻性,也有助於提昇呼吸醫療監護的普及與效率。This thesis proposes some analytic methods to respiratoryphysio logicalsignals. The goal of this thesis is to createfeatures of respiratory signals and apply them to the clinicaldiagnosis and monitoring care system of respiratory systemdiseases. Being dif ferent with ECG signals, respiratorywaveforms exist no unique s hape. Even for unique subject,respiratory signals acquisited in different situation and timewill differ each others. So, the po ssible approach to identifyrespiratory waveforms with different respiratory diseases isusing proper respiratory features to ind icate thecharacteristics of waveforms. This thesis extracts res piratoryfeatures from time domain, frequency domain and systemmodeling. The purpose of feature extraction is to takeadvantage of the difference of respiratory features fordifferent diseases to achieve the recognition andclassification of respiratory dis eases. This thesisresearches a general and integrated solution of respiratorysignal processing. First, through choice of respi ratorysensors, selection and design of A/D and amplifier interf acecircuits, and conversion of monitoring care data, we can acquisit respiratory data. Then, the proposed methods are usedto extract respiratory features. The extracted features areused to train artificial neural network to activate theirclassification ability. Finally, this thesis proposes someclinical application s and conclusions based on result ofsimulations.

Metrics

1 Record Views

Details

Logo image