Computational Intelligence and Healthcare Informatics

5Classification of Heart Sound Signals Using Time-Frequency Image Texture Features

Sujata Vyas¹, Mukesh D. Patil¹ and Gajanan K. Birajdar²

¹Department of Electronics and Telecommunication Engineering, Ramrao Adik Institute of Technology, Nerul, Navi Mumbai, Maharashtra, India

²Department of Electronics Engineering, Ramrao Adik Institute of Technology, Nerul, Navi Mumbai, Maharashtra, India

Abstract

This book chapter presents an approach for heart sound classification based on time-frequency image texture feature and support vector machine classifier. Firstly, the spectrogram image is generated to create a time-frequency representation of small frames from an input speech sample. The spectrogram visual representation provides finest details over resolutions which can be tuned to capture heart sound detailed information. We then acquire a local binary pattern (LBP), and its derivatives like complete local binary pattern (CLBP), dense local binary pattern (dense CLBP), local directional texture pattern (LDTP), and Weber local descriptor (WLD), which is better suited to extract features from heart sound signal spectrogram images. To select prominent features and to discard redundant features, a chaotic moth flame optimization algorithm is employed. Experimental results yield a 5.2% improvement on the PhysioNet 2016 Database compared to the short-time Fourier transform method with LBP. Fusing Mel-spectrogram and IIR constant-Q transform method in pre-processing, a classification ...