An improved method to detect coronary artery disease using phonocardiogram signals in noisy environment Akanksha Pathak a,⇑ , Pranab Samanta a , Kayapanda Mandana b , Goutam Saha a a Department of Electronics and Electrical Communication Engineering, Indian Institute of Technology, Kharagpur, Kharagpur 721302, India b Department of Cardiology, Fortis Healthcare, Kolkata 700107, India article info Article history: Received 23 May 2019 Received in revised form 23 January 2020 Accepted 31 January 2020 Keywords: Coronary artery disease (CAD) Cross power spectral density (CPSD) Environmental noise Multichannel Phonocardiogram (PCG) Subband features abstract Identification of coronary artery disease (CAD) from phonocardiogram (PCG) signal is a low signal to noise ratio (SNR) problem. This study proposes a PCG based CAD detection system robust against the environmental noise that does not require additional reference signals for noise acquisition and PCG seg- mentation. Here, the experiments are conducted on 40 CAD and 40 normal subjects. PCG signals are recorded from a multichannel data acquisition system from four auscultation sites on the left anterior chest. While heart sounds are propagated to different auscultation sites with a certain delay, the ambient noise appearing at microphone array are not mutually time-lagged. Thus, we propose to use the imagi- nary part of cross power spectral density (ICPSD) to capture the spectrum of heart sounds as it is unre- sponsive to zero time-lagged signals. Subband based spectral features obtained from ICPSD are classified in a machine learning framework. The performance of the system is studied in the presence of babble, vehicle and white noise in which useful information were extracted from both systolic and diastolic phases of cardiac cycle. The proposed method achieves accuracy, sensitivity and specificity of 74.98%, 76.50% and 73.46%, respectively in absence of ambient noise for k-fold (k ¼ 5) cross-validation. The accu- racy for 0 dB SNR in presence of white, babble and vehicle noise were 71:13%; 66:47% and 69:60%, respectively. The proposed method was found to be superior in CAD classification when compared with existing noise removal based approach. The present work shows the potential of developing a PCG-based multichannel CAD detection system as an affordable point of care device for real-life use, where a certain amount of ambient noise is expected. Ó 2020 Elsevier Ltd. All rights reserved. 1. Introduction The development of signal processing and machine learning algorithms has brought a surge in mining diagnostic information from numerous biosignals generated by the human body. The biosignals are markers for the physiological traits of the human organs [1] and can provide a noninvasive, easy-to-use, affordable and low-risk diagnosis. Amidst various biosignals, phonocardio- gram (PCG), an electronic record of heart sounds acquired using auscultation from the chest, can play an important role to diagnose chronic cardiovascular diseases such as coronary artery disease (CAD). CAD mainly occurs due to atherosclerosis [2]. It is defined as ‘‘variable combination of changes of the intima of arteries con- sisting of focal accumulation of lipids, complex carbohydrates, blood and blood products, fibrous tissues and calcium deposits and associated with medial changes” [2]. These depositions, collec- tively referred to as plaque, narrows the internal diameter of the artery, subsequently causing a reduction in oxygenated blood sup- ply to heart muscles. In spite of these complexities, CAD due to atherosclerosis is a preventable disease. Thus, its early detection can stop the progression to its severe stages such as, ischemic heart disease (IHD), myocardial infarction (MI) and congestive heart failure (CHF) [2]. The movement of bloodstream in healthy arteries is stream- lined and flows at critical velocity [3]. Stenosed arteries cause blood flow above critical velocity, leading to turbulent flow. The turbulent flow may or may not generate murmurs [4], while streamlined flow remains silent [3]. In addition to murmurs, steno- sis sets narrowed orifices, endothelial surfaces and walls of adjoin- ing tissues into vibration [5,6]. These murmurs and vibrations are https://doi.org/10.1016/j.apacoust.2020.107242 0003-682X/Ó 2020 Elsevier Ltd. All rights reserved. Abbreviations: CAD, Coronary artery disease; ICPSD, Imaginary part of cross power spectral density; MA, Midaxillary; MCPSD, Magnitude of cross power spectral density; MIT, Mitral; PCG, Phonocardiogram; PUL, Pulmonary; TRI, Tricuspid. ⇑ Corresponding author. E-mail address: akanksha28.iitkgp@gmail.com (A. Pathak). Applied Acoustics 164 (2020) 107242 Contents lists available at ScienceDirect Applied Acoustics journal homepage: www.elsevier.com/locate/apacoust