Heart Sounds Classification Using Frequency Features with Deep Learning Approaches

Abstract

Cardiovascular diseases are diseases that attack the heart or the blood vessels. Since the heart is the body’s most important organ, it is crucial to spot anomalies in its functioning. Many researchers have examined various automated categorization techniques for heart sounds, which has been a topic of interest for decades. Thanks to machine learning techniques, large datasets of heart sounds may now be automatically evaluated. Recent studies have used support vector machines, and neural networks to categorize heart sounds more accurately. The highest reported accuracy of 0.99 was attained by a Convolutional Neural Network-based model that utilized the short-time Fourier transform and Mel-frequency Cepstral coefficients as features. However, the worth of the input signal and the variation in heart sounds between people can have an impact on how accurate these algorithms are. This paper exhibits the potential of machine learning models and deep learning models in classifying heart sounds. An accuracy of 0.8884 has been reached by applying a Convolutional Neural Network model on the spectrogram feature extracted from the heart sounds.