A novel method for discrimination between innocent and pathological heart murmursby Arash Gharehbaghi, Magnus Borga, Birgitta Janerot Sjöberg, Per Ask

Medical Engineering & Physics


Biophysics / Biomedical Engineering




Can Cardiologists Distinguish Innocent from Pathologic Murmurs in Neonates?

Andrew S. Mackie, Luc C. Jutras, Adrian B. Dancea, Charles V. Rohlicek, Robert Platt, Marie J. Béland

The American Academy of Fine Arts in Rome

Adapted for Brush and Pencil

The patient with a heart murmur



Medical Engineering and Physics 37 (2015) 674?682

Contents lists available at ScienceDirect

Medical Engineer journal homepage: www.elsev en rot S , Link? gineer and Te meth or ma moreA novel method for discrimination betwe heart murmurs

Arash Gharehbaghia,?, Magnus Borgaa,b, Birgitta Jane a Physiological Measurements, Department of Biomedical Engineering, Link?ping University b Center for Medical Image Science and Visualization (CMIV), Department of Biomedical En cDivision of Medical Imaging and Technology, Department of Clinical Science, Intervention dDepartment of Clinical Physiology, Karolinska University Hospital, Stockholm, Sweden e School of Technology and Health, KTH Royal Institute of Technology, Stockholm, Sweden a r t i c l e i n f o

Article history:

Received 11 March 2014

Revised 18 November 2014 a b s t r a c t

This paper presents a novel growing time support vect murmurs (IM) by putting

Accepted 25 April 2015


Growing-time support vector machine

Support vector machine


Heart murmurs

Innocent murmurs systolic phase. Individuals with analysis, taking the normal indi to the similarity of its murmur time windows, the performance (SVM), using repeated random is found to be 88%/86% for the G

GTSVM significantly improves p 1. Introduction

Heart sound auscultation is a powerful technique for screening patients with heart diseases for cardiac examinations. It is often based on the subjective judgments of the temporal and spectral characteristics of the heart sounds and murmurs. Previous studies showed that the joint time-frequency contents of heart murmurs convey diagnostic information about mechanical activities of heart [1?3]. However, physiological restrictions of human auditory system (i.e. masking effect) along with the spectral properties of the cardiac sounds and murmurs (concentrated down to the human auditory threshold), make the subjective assessment a complicated and sometimes problematic task [3,4]. As a result, the number of healthy referrals to the hospitals for echocardiographic investigation is substantial (as many as 70%), showing the fact that accuracy of the conventional auscultation is inadequate [5,6]. Normally, a phonocardiogram (PCG) is composed of two basic sounds in each cardiac cycle; the first heart sound (S1) and the second heart sound (S2) [7]. Heart murmurs are extra sounds, heard either in systolic phase (from S1 to S2) or in the diastolic one (from S2 to S1) [7]. Systolic murmurs ? Corresponding author. Tel.: +46 13 286754; fax: +46 13 101902.

E-mail address: arash.ghareh.baghi@liu.se, arash_gharahbaghi@yahoo.com (A. Gharehbaghi). c p m c fi a c l d g i m c s

I s s b p b m t t http://dx.doi.org/10.1016/j.medengphy.2015.04.013 1350-4533/? 2015 IPEM. Published by Elsevier Ltd. All rights reserved.ing and Physics ier.com/locate/medengphy innocent and pathological j?bergc,d,e, Per Aska ping, Sweden ing, Link?ping University, Link?ping, Sweden chnology, Karolinska Institutet, Stockholm, Sweden od for discrimination between innocent and pathologicalmurmurs using the chine (GTSVM). The proposed method is tailored for characterizing innocent emphasis on the early parts of the signal as IMs are often heard in earlymild to severe aortic stenosis (AS) and IM are the two groups subjected to viduals with no murmur (NM) as the control group. The AS is selected due to IM, particularly in mild cases. To investigate the effect of the growing of the GTSVM is compared to that of a conventional support vector machine sub-sampling method. The mean value of the classification rate/sensitivity

TSVM and 84%/83% for the SVM. The statistical evaluations show that the erformance of the classification as compared to the SVM. ? 2015 IPEM. Published by Elsevier Ltd. All rights reserved. onstitute the main group of the extra sounds and can be either a athological or non-pathological sign. Diastolic murmurs (the murur which is heard only in diastole) are usually assumed pathologial, albeit a diastolic third or even forth heart sound may be a normal nding, especially in children [8]. Innocent systolic murmurs (IM) is group of the non-pathological sounds, heard in a great majority of hildren (up to 70%) [6,9]. IM is manifested as a systolic murmur with ow to medium intensity, and ends soon after the middle of systolic uration (mid systole) [10,11]. However, some of the heart patholoies could have similar manifestation (i.e. mild aortic stenosis) which nvolves further complexity in correct detection of the pathological urmurs (PM) even for expert physicians [7]. The onset of IM coinides with early systolic part which is sometimes similar to PM. Fig. 1 hows a typical cardiac cycle, selected from the three conditions: PM,

M and no murmurs (NM).

As seen in Fig. 1, the amplitude of the murmurs does not necesarily show noticeable discriminations.

Considerable research has been directed toward automatic creening of the pathological heart sounds and murmurs [2,12?15], ut not so to the discrimination between IM and PM with mild athologies [6,16,17]. Non-stationary property of the murmurs is a ig obstacle for developing an automatic algorithm for heart murur classification. The non-stationarity is also observed as the beat o beat variation. These facts make statistical analysis and optimizaion essential needs for a robust classification.

A. Gharehbaghi et al. /Medical Engineering and Physics 37 (2015) 674?682 675 ocent f v c s w f s l m c t t c d t a 2 e

T s t t t o s i q w iFig. 1. A complete cardiac cycle of a normal (bottom), inn

This paper proposes a novel artificial intelligence-based methodor discrimination between IM and PM using growing time support ector machine (GTSVM). Although the method is essentially appliable on any segment of the PCG, however, we selected the part of ystolic duration between S1 and S2 as the input signal to themethod, here the IM is commonly heard. Details of the automatic algorithms or PCG segmentation could be found in [18,19]. The method employs tatistical techniques both in training and optimization to elaborately earn non-stationary properties of the systolic murmur. The perforance of the method is validated using four databases, altogether onsisting of the PM, IM and NM conditions. The PM contains mild o severe aortic stenosis (AS). The main motivation of selecting AS is hat the murmur of AS (in mild cases) is sometimes similar to and onfused by IM (see Fig. 1).