- 學位論文
利用小波轉換以及模糊神經網路進行震盪血壓波型之病徵分類
Blood Pressure Classification Using Wavelet And Fuzzy Neural Network
摘要
心血管疾病一直是威脅人類性命的主要原因。血壓波型包含了心臟與血管生理狀況的訊息，所以本研究提議探討血壓波型上的特徵，期望能找出與不同程度心血管疾病相依性高的血壓波型特徵來進行病症分類，以提升分類的準確性。 本研究共收集了61位受測者，其中包含22位正常人、20位冠心症病患和19位周邊血管阻塞病患的血壓波型。本研究擷取每位受測者於收縮壓、舒張壓和平均血壓這三個壓力下，血壓訊號上的波型特徵，共68個。其中利用小波轉換偵測的重搏切點因為偵測誤差過大，而被排除。其他特徵經由篩選，最後獲得28個最佳的特徵，做為模糊神經網路的輸入。 首先本研究利用模糊神經網路進行正常人和周邊血管阻塞病患之分類，隨機選擇2/3的受測者進行網路訓練，其他的1/3的受測者進行測試，結果顯示分類的準確度為100%。顯示出正常人與周邊血管阻塞之病患的這28個特徵具有高度的識別能力。另一方面正常人、冠心症和周邊血管阻塞的病患的分類，結果顯示最高可以達到77.78%的準確率。之所以無法達到較好的分類結果也許是因為收集到的冠心症的病患都是已經做過心臟支架手術，其血管狀況趨近於正常人而造成分類上的錯誤。如果能收取未進行支架手術前的冠心症病患的訊號則更能突顯此類別的病患與正常人之間的差異。 本研究結果顯示利用血壓波型上的特徵，經模糊神經網路能有效的分類病症。
並列摘要
Cardiovascular disease always is a main threat to human health. On the other hand, blood pressure waveform contains physiological information of heart and blood vessel. Thus, this study proposes to investigate the characteristic features of blood pressure waveform in order to find features that are correlated with different degrees of cardiovascular disease to improve the classification accuracy. This study recruited 61 subjects, 22 of them are normal subjects, 19 are PAD subjects, and 20 CAD subjects. Three pressure waveforms were collected at systolic, diastolic and mean blood pressures for every subject. A total of 68 features were extracted from these waveforms. However, features obtained using wavelet were excluded due to exceed error. Other features were further screened and 28 features were obtained for fuzzy neural network analysis. First, this study uses FNN to classify normal and PAD subjects. Two third of the subjects were selected randomly for network training and the rest of the subjects for testing. The results indicate that the accuracy of classification is 100%. This indicates that these 28 features have clear capability in classification. On the other hand, the best accuracy rate is 77.78% for normal, PAD and CAD classification. The reason for not getting better result may be due to the criteria in CAD screening was subjects who have stem operation. The properties of their blood vessel are similar to those of normal subject. If we can obtain signal from subjects before stem operation, the difference may be much more significant. In conclusion, the results of this study prove that, by using fuzzy neural network, features in blood pressure waveform are capable in disease classification.