Biomedical Signal Processing and Control 36 (2017) 11–19 Contents lists available at ScienceDirect Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc Automatic cardiac phase detection of mitral and aortic valves stenosis and regurgitation via localization of active valves Azra Saeidi a , Farshad Almasganj a,∗ , Maryam Shojaeifard b a Biomedical Engineering Department, Amirkabir University of Technology, Tehran, Iran b Rajaie Cardiovascular Medical & Research Center, Tehran, Iran a r t i c l e i n f o Article history: Received 29 December 2016 Received in revised form 4 March 2017 Accepted 22 March 2017 Keywords: Cardiac phase detection Mitral and Aortic stenosis Mitral and aortic regurgitation Heart valves locations RAP-group delayed MUSIC a b s t r a c t Heart auscultation is a primary way of heart diagnosis and monitoring. A critical step of heart sound analysis is dividing the sound into its basic phases, known as S1, systole, S2 and diastole. Heart sound phase detection becomes a challenging issue when more than one heart valve problems are simultane- ously existed. In this paper, a powerful approach is proposed that properly determine phases of heart sounds with Mitral and Aortic stenosis and Regurgitation murmurs. In this approach, heart valves spatial- temporal information investigated by a multiple channel data recording system is employed to determine different heart phases. Since heart valves are very near to each other and are located in the reverberant environment of chest, according to its anatomy, the group-delay information of Recursively Applied and Projected-Multiple Signal Classification spectra (RAP-MUSIC) is utilized to localize active heart valves. The proposed segmentation algorithm is applied to some normal and abnormal (mitral and aortic valve malfunctions) heart sounds recorded by a rectangular microphone array consisted of six sensors. To evaluate the benefits of the proposed method, it is compared with the basic and Tunable Q wavelet and S-transform based segmentation methods, by conducting some proper experiments. The obtained results show that the proposed algorithm is considerably superior to the mentioned methods The average pos- itive predictive value, sensitivity and accuracy results from the proposed segmentation algorithm are 97.4%, 95.5%, and 93%, respectively. © 2017 Elsevier Ltd. All rights reserved. 1. Introduction Heart sound has a limited frequency bandwidth, within a range of 10–750 Hz [1]. This low intensity, complex and non-stationary audio signal contains a lot of physiological and pathological infor- mation of heart valve which is helpful for diagnosing heart valve abnormalities [1,2]. Heart auscultation is a simple, low-cost and non-invasive heart monitoring method, but it is highly dependent on the specialist skills in heart diagnosing. In this respect, auto- matic heart sound analysis can be considered as an auxiliary heart valve disorder classification method. Heart muscle vibration produces four basic heart phases: S1, systole, S2, and diastole. The S1 phase is composed of the closure of the Mitral and the Tricuspid valves. The next phase is called sys- tole where the ventricles contract and atrio-ventricular (AV) canal is closed and heart chambers eject blood. The S2 phase starts due to the closure of the Aortic and Pulmonary valves. The last phase ∗ Corresponding author. E-mail address: almas@aut.ac.ir (F. Almasganj). of heart in which the ventricles refills with blood is called Diastole. Only S1 and S2, could be clearly heard in normal heart sounds. In abnormal heart sounds, due to turbulent flows through stenosis valves or backward flows through regurgitation valves, some extra sounds, called murmurs, are produced during the systole and dias- tole heart phases [3]. Each murmur is characterized by its quality of sound, shapes and frequency pitches. Common murmurs Aortic Stenosis (AS) and Mitral Regurgitation (MR) are occurred in systolic phase. While Aortic Insufficiency (AI), and Mitral Stenosis (MS) are diastolic murmur. The Phonocardiography (PCG) segmentation is the most impor- tant step in heart sound analysis in which the heart sound is labeled into S1, systole, S2, and diastole periods. There are two categories of heart sound segmentation methods: supervised and unsuper- vised. In the supervised methods, a reference Echocardiography (ECG) signal is needed to be aligned with the newly recorded signal [4,5]. The supervised ECG based segmentation methods typically have high performances [6]; however, this type of segmentation may not be convenient for long term monitoring purposes, due to high demand of extra memory space for the ECG and PCG signals processing, higher computational cost and power consumption. In http://dx.doi.org/10.1016/j.bspc.2017.03.005 1746-8094/© 2017 Elsevier Ltd. All rights reserved.