Phonocardiography signal processing for automatic diagnosis of ventricular septal defect in newborns and children

• Published in: 2017 9th International Conference on Computational Intelligence and Communication Networks (CICN) pp. 62 - 66
• Main Authors: Ghaffari, Milad, Ashourian, Mohsen, Ince, Erhan A., Demirel, Hasan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-09-2017
• Subjects: cardiac murmurs; Feature extraction; Frequency-domain analysis; Graphical user interfaces; Heart; Pediatrics; Phonocardiography; phonocardiography signal processing; Signal processing; ventricular septal defect
• ISSN: 2472-7555
• DOI: 10.1109/CICN.2017.8319357

In medical literature auscultation of heart sounds is an important skill for diagnosing cardiac cases, but is associated with many difficulties. In this study, a phonocardiography (PCG) based system is presented that could aid automatic analysis of heart sounds and diagnosis of ventricular septal defect (VSD) in newborns and children. Since the degree of the septal defect can be determined based on the diameter of defect (small > 3 mm to < 6 mm, moderate > 6 mm to < 12 mm, and large > 12 mm), our system would also report the septal defect's diameter. The proposed phonocardiography system for detecting and diagnosing the congenital heart diseases is simple, inexpensive and non-invasive which can be an alternative method for diagnosing VSD instead of echocardiography. In this study, recorded cardiac sounds of 22 newborns aged between 6 months to 2 years who were previously proved to have various cardiac diseases were used. Digital signal processing techniques such as short-time Fourier transform (STFT), segmentation and autocorrelation, Mel Frequency Cepstral Coefficients (MFCC) and their derivatives were used to extract features and these features were then classified using the K-Nearest Neighbors algorithm (KNN). A brief analysis of the results showed that for 93.2% of the test cases the proposed phonocardiography based system would correctly diagnose the recordings and the average defect diameter deviation was 6.79%.