TY - JOUR
T1 - Analysis and methods to test classification of normal and pathological heart sound signals
AU - Hendradi, Rimuljo
AU - Arifin, Achmad
AU - Shida, Hiro
AU - Gunawan, Suhendar
AU - Purnomo, Mauridhi Hery
AU - Hasegawa, Hideyuki
AU - Kanai, Hiroshi
N1 - Publisher Copyright:
© 2005 - 2016 JATIT & LLS. All rights reserved.
PY - 2016/8/15
Y1 - 2016/8/15
N2 - An acute shortage of cardiologists and many rural clinics were run by nurses in Indonesia. We proposed to develop of a screening technique based on artificial intelligence that classifies of normal and pathological heart sound signals of human subjects due to signs important and symptoms for heart diagnosis based on knowledge of auscultation experts. Heart sound signal analysis system consisted of three stages. Firstly, preprocessing. Secondly, feature extraction with respect to the cardiac cycle based on wavelet analysis to differentiate normal and pathological heart sounds. Feature reduction using PCA was also carried out to reduce the dimension of the heart sound feature vectors for classification. Thirdly, three classifiers: ANN MLP-BP, FCM clustering and HCM clustering to classify normal, systolic murmur, diastolic murmur, and continuous murmur, respectively. The performance of each classifier was evaluated with statistical validation method. From our experimental results, the three classifiers that showed significant potential in their use as an alternative diagnostic tool were compared. The ANN achieved the best performance as an automated classifier rather than FCM and HCM methods. Its performance was 100% for sensitivity, specificity, and accuracy, respectively, of input 20,000 features. Furthermore, for input 300 features, the performance was 98.90%, 99.37%, and 99.03% for sensitivity, specificity, and accuracy, respectively. The heart sound signal analysis system was suitable to classify of normal and pathological cases. The proposed method was considered very important for objective screening and very useful as an alternative diagnostic tool that complies with the requirements for rural clinics. We hoped that the method would be beneficial in study of auscultatory technique for medical students. Surrogate data modeling of pathological heart sounds signals as an alternative tool of the heart sound simulator and for classification purpose was further study.
AB - An acute shortage of cardiologists and many rural clinics were run by nurses in Indonesia. We proposed to develop of a screening technique based on artificial intelligence that classifies of normal and pathological heart sound signals of human subjects due to signs important and symptoms for heart diagnosis based on knowledge of auscultation experts. Heart sound signal analysis system consisted of three stages. Firstly, preprocessing. Secondly, feature extraction with respect to the cardiac cycle based on wavelet analysis to differentiate normal and pathological heart sounds. Feature reduction using PCA was also carried out to reduce the dimension of the heart sound feature vectors for classification. Thirdly, three classifiers: ANN MLP-BP, FCM clustering and HCM clustering to classify normal, systolic murmur, diastolic murmur, and continuous murmur, respectively. The performance of each classifier was evaluated with statistical validation method. From our experimental results, the three classifiers that showed significant potential in their use as an alternative diagnostic tool were compared. The ANN achieved the best performance as an automated classifier rather than FCM and HCM methods. Its performance was 100% for sensitivity, specificity, and accuracy, respectively, of input 20,000 features. Furthermore, for input 300 features, the performance was 98.90%, 99.37%, and 99.03% for sensitivity, specificity, and accuracy, respectively. The heart sound signal analysis system was suitable to classify of normal and pathological cases. The proposed method was considered very important for objective screening and very useful as an alternative diagnostic tool that complies with the requirements for rural clinics. We hoped that the method would be beneficial in study of auscultatory technique for medical students. Surrogate data modeling of pathological heart sounds signals as an alternative tool of the heart sound simulator and for classification purpose was further study.
KW - Artificial Neural Network Multilayer Perceptron Back Propagation (ANN MLP-BP)
KW - Fuzzy C-Means (FCM) Clustering
KW - Hard CMeans (HCM) Clustering
KW - Principal Component Analysis (PCA)
KW - Wavelet Analysis
UR - http://www.scopus.com/inward/record.url?scp=84982156055&partnerID=8YFLogxK
M3 - Article
AN - SCOPUS:84982156055
SN - 1992-8645
VL - 90
SP - 222
EP - 236
JO - Journal of Theoretical and Applied Information Technology
JF - Journal of Theoretical and Applied Information Technology
IS - 1
ER -