RETRACTED ARTICLE: Classification of unsegmented phonocardiogram signal using scalogram and deep learning

• Published in: Soft computing (Berlin, Germany) Germany), 2023-09, Vol.27 (17), p.12677-12689
• Main Authors: Devi, Kshetrimayum Merina, Chanu, Maibam Mangalleibi, Singh, Ngangbam Herojit, Singh, Khumanthem Manglem
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Cardiovascular disease; Classification; Computational Intelligence; Continuous wavelet transform; Control; Datasets; Deep learning; Electrocardiography; Engineering; Focus; Fourier transforms; Heart diseases; Heart failure; Machine learning; Mathematical Logic and Foundations; Mechatronics; Neural networks; Performance measurement; Phonocardiography; Robotics; Sound; Success; Support vector machines
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-023-08834-1

In many parts of the globe, heart disease or cardiovascular disease is a leading cause of mortality. It is a group of conditions that includes heart failure and other cardiac problems. In the USA, there are an estimated 356,000 cases of hospital cardiac arrest every year, 90% of which are fatal, as reported by the American Heart Association (Tsao et al. in Circulation https://doi.org/10.1161/CIR.0000000000001052 , 2022). However, the survival rate may be improved by routine monitoring of cardiac function in order to detect the illness at an early stage. Many formerly intractable medical issues have made significant progress because of the combination of computer technology and biomedical research. Automatic prediction is made possible by the examination of previously learned data in machine learning. In this research, we utilize the Physionet Cinc/C Challenge 2016 Phonocardiogram dataset. Classification models like K-nearest neighbour (KNN), support vector machine (SVM), and AlexNet are used with the features retrieved using wavelet scattering transform and continuous wavelet transform. The primary objective of this work is to provide a very effective method for early diagnosis of cardiovascular disorders, which will help clinical professionals better spot heart issues in their patients. Sensitivity, Specificity and other performance metrics are used to assess the effectiveness of the various categorization models. KNN, SVM, and AlexNet classifiers all achieve good accuracy with AlexNet outperforms the former two giving an accuracy of 98.11%, with a sensitivity of 96.20% and a specificity of 98.73%. Our suggested model’s output has the potential to contribute significantly to the categorization of cardiac sounds.