Limits...
A system for heart sounds classification.

Redlarski G, Gradolewski D, Palkowski A - PLoS ONE (2014)

Bottom Line: The future of quick and efficient disease diagnosis lays in the development of reliable non-invasive methods.Thanks to the advancement in technology, the quality of phonocardiography signals is no longer an issue.The respective system is compared with four different major classification methods, proving its reliability.

View Article: PubMed Central - PubMed

Affiliation: Department of Mechatronics and High Voltage Engineering, Gdansk University of Technology, Gdansk, Poland.

ABSTRACT
The future of quick and efficient disease diagnosis lays in the development of reliable non-invasive methods. As for the cardiac diseases - one of the major causes of death around the globe - a concept of an electronic stethoscope equipped with an automatic heart tone identification system appears to be the best solution. Thanks to the advancement in technology, the quality of phonocardiography signals is no longer an issue. However, appropriate algorithms for auto-diagnosis systems of heart diseases that could be capable of distinguishing most of known pathological states have not been yet developed. The main issue is non-stationary character of phonocardiography signals as well as a wide range of distinguishable pathological heart sounds. In this paper a new heart sound classification technique, which might find use in medical diagnostic systems, is presented. It is shown that by combining Linear Predictive Coding coefficients, used for future extraction, with a classifier built upon combining Support Vector Machine and Modified Cuckoo Search algorithm, an improvement in performance of the diagnostic system, in terms of accuracy, complexity and range of distinguishable heart sounds, can be made. The developed system achieved accuracy above 93% for all considered cases including simultaneous identification of twelve different heart sound classes. The respective system is compared with four different major classification methods, proving its reliability.

Show MeSH

Related in: MedlinePlus

Comparison of classification results for all test groups.A comparison of accuracy of all tested methods (being: ANN – Artificial Neural Network, SVM-poly – Support Vector Machine with polynomial kernel function, SVM-rbf – Support Vector Machine with radial basis kernel function, SVM-quad – Support Vector Machine with quadratic kernel function, SVM-MCS-ca – Support Vector Machine with the Modified Cuckoo Search optimizer and classification accuracy fitness function) for a different number of recognizable classes. The SVM-MCS-ce method shows overall the best quality of classification.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4231067&req=5

pone-0112673-g008: Comparison of classification results for all test groups.A comparison of accuracy of all tested methods (being: ANN – Artificial Neural Network, SVM-poly – Support Vector Machine with polynomial kernel function, SVM-rbf – Support Vector Machine with radial basis kernel function, SVM-quad – Support Vector Machine with quadratic kernel function, SVM-MCS-ca – Support Vector Machine with the Modified Cuckoo Search optimizer and classification accuracy fitness function) for a different number of recognizable classes. The SVM-MCS-ce method shows overall the best quality of classification.

Mentions: Tables 3 and 4 present classification results for a different number of distinguishable classes. A graphic representation of those results can be seen in Figure 8. The Support Vector Machine with the Modified Cuckoo Search optimizer and classification efficiency fitness function classifier again has the best efficiency. The most notable result is its capability to classify 12 heart sounds at the same time with a rate of 93.23%. Its scores are never less than 92%, which compared to the other methods, presents outstanding performance. The only cases when it performs worse are the first four tests of the ANN, when the ANN achieved a perfect rate. However, for more than six classes its performance dropped significantly, which might indicate that medical examination using this classifier could be very uncertain. The performance of the SVM classifiers rises with an increase in the number of classes, reaching its maximum of 92.36% with a polynomial kernel function. It should be noted that this result is worse even when compared to the SVM-MCS system using support vector number objective function. This poor performance in the case of a small number of classes can be explained by over-fitting in the training phase caused by maladjustment of the SVM parameters.


A system for heart sounds classification.

Redlarski G, Gradolewski D, Palkowski A - PLoS ONE (2014)

Comparison of classification results for all test groups.A comparison of accuracy of all tested methods (being: ANN – Artificial Neural Network, SVM-poly – Support Vector Machine with polynomial kernel function, SVM-rbf – Support Vector Machine with radial basis kernel function, SVM-quad – Support Vector Machine with quadratic kernel function, SVM-MCS-ca – Support Vector Machine with the Modified Cuckoo Search optimizer and classification accuracy fitness function) for a different number of recognizable classes. The SVM-MCS-ce method shows overall the best quality of classification.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4231067&req=5

pone-0112673-g008: Comparison of classification results for all test groups.A comparison of accuracy of all tested methods (being: ANN – Artificial Neural Network, SVM-poly – Support Vector Machine with polynomial kernel function, SVM-rbf – Support Vector Machine with radial basis kernel function, SVM-quad – Support Vector Machine with quadratic kernel function, SVM-MCS-ca – Support Vector Machine with the Modified Cuckoo Search optimizer and classification accuracy fitness function) for a different number of recognizable classes. The SVM-MCS-ce method shows overall the best quality of classification.
Mentions: Tables 3 and 4 present classification results for a different number of distinguishable classes. A graphic representation of those results can be seen in Figure 8. The Support Vector Machine with the Modified Cuckoo Search optimizer and classification efficiency fitness function classifier again has the best efficiency. The most notable result is its capability to classify 12 heart sounds at the same time with a rate of 93.23%. Its scores are never less than 92%, which compared to the other methods, presents outstanding performance. The only cases when it performs worse are the first four tests of the ANN, when the ANN achieved a perfect rate. However, for more than six classes its performance dropped significantly, which might indicate that medical examination using this classifier could be very uncertain. The performance of the SVM classifiers rises with an increase in the number of classes, reaching its maximum of 92.36% with a polynomial kernel function. It should be noted that this result is worse even when compared to the SVM-MCS system using support vector number objective function. This poor performance in the case of a small number of classes can be explained by over-fitting in the training phase caused by maladjustment of the SVM parameters.

Bottom Line: The future of quick and efficient disease diagnosis lays in the development of reliable non-invasive methods.Thanks to the advancement in technology, the quality of phonocardiography signals is no longer an issue.The respective system is compared with four different major classification methods, proving its reliability.

View Article: PubMed Central - PubMed

Affiliation: Department of Mechatronics and High Voltage Engineering, Gdansk University of Technology, Gdansk, Poland.

ABSTRACT
The future of quick and efficient disease diagnosis lays in the development of reliable non-invasive methods. As for the cardiac diseases - one of the major causes of death around the globe - a concept of an electronic stethoscope equipped with an automatic heart tone identification system appears to be the best solution. Thanks to the advancement in technology, the quality of phonocardiography signals is no longer an issue. However, appropriate algorithms for auto-diagnosis systems of heart diseases that could be capable of distinguishing most of known pathological states have not been yet developed. The main issue is non-stationary character of phonocardiography signals as well as a wide range of distinguishable pathological heart sounds. In this paper a new heart sound classification technique, which might find use in medical diagnostic systems, is presented. It is shown that by combining Linear Predictive Coding coefficients, used for future extraction, with a classifier built upon combining Support Vector Machine and Modified Cuckoo Search algorithm, an improvement in performance of the diagnostic system, in terms of accuracy, complexity and range of distinguishable heart sounds, can be made. The developed system achieved accuracy above 93% for all considered cases including simultaneous identification of twelve different heart sound classes. The respective system is compared with four different major classification methods, proving its reliability.

Show MeSH
Related in: MedlinePlus