Although Dr. Laennec's invention, the stethoscope, has been in clinical use for more than 180 years, and electronic stethoscopes with variable amplification gain have been available for over 80 years, it is still difficult to understand auscultation findings [1-3]. The phonocardiogram, first developed in 1894, visualizes auscultatory signals [2,4,5]. The spectral phonocardiogram has proven to be a reliable tool that gives information of whether or not the murmur is pathological. Based on earlier studies and clinical observations, it has been assumed that pathological murmurs involve sounds of higher frequency [2,5]. Phonocardiography and electronic stethoscopy attempt to improve the diagnostic accuracy of cardiac auscultation. In the most recent studies, digital acoustic analysis has demonstrated the validity of these methods [6-11]. Since the 1980's, phonocardiographic research activity had decreased due to the improvements of echocardiography, which yields more visual information. During the past few years, however, the improvements of personal computers have made it possible to design new low-cost, high-quality phonocardiographic devices [12-17]. Spectral phonocardiography emulates the ear and may be ideal for teaching clinical stethoscopy [4]. The phono-spectrogram combines traditional phonocardiogram with time-frequency distribution presentation of the signal. The spectrogram was introduced for heart sound analysis as early as 1955 by McKusik et al, but was afterwards almost forgotten [4,11].

The first evaluation of children's heart murmur is one of the basic tasks of welfare clinic practitioners and school medical officers. However, based on their frequently limited auscultation experience, they might not be able to recognize the innocence of a heart murmur. Even the auscultation skills of pediatric residents have been found to be suboptimal [18-21]. Several healthy children are referred to pediatric cardiologists or for echocardiography. Parental anxiety may also be a reason to refer the patient for unnecessary examinations, and it is important to recognize an innocent murmur as soon as it is found [22-25]. Auscultation training will naturally improve practitioners' listening skills [26-28].

Previous studies have shown that pediatric cardiologists, based on clinical examination, can differentiate innocent from pathological murmurs with high sensitivity (82...92 %) and specificity (76...99%) [29-33]. Even better results have been attained by using advanced signal processing and pattern recognition tools. Artificial neural network-based screening is reported to have 100% sensitivity and specificity [13,14].

The capability of a doctor or a computerized system to differentiate between pathological and innocent murmurs depends on the quality of the pathological murmurs. It is easy to recognize loud murmurs as pathological by means of clinical or digital analysis [9]. Serious defects are seldom missed, and the challenge is actually to detect mild defects. Small muscular VSD, mild PS or AS causes no hemodynamic harm, but they may require endocarditis prophylaxis. ASD secundum may also be hard to diagnose only by auscultation [34-38].

The purpose of this article was to demonstrate the capabilities of a modern digital system for phonocardiographic recording and analysis and to evaluate its potential for differentiating the characteristic features of heart murmurs in children. Another goal was to investigate how the numerical features measured from a phono-spectrogram can be used to distinguish innocent systolic murmurs from pathological murmurs, which are generally hard to identify in a routine clinical examination. This article additionally concludes more than ten years' experience of studying heart sounds.

Table 1 Recorded heart sounds

The examiner heard the sounds through earphones and monitored the signals on the computer screen. Thus, the recording was completed in almost the same time as traditional auscultation. Some additional effort was needed to enter the patient information. The new generation commercial electronic stethoscopes were also compatible with this recording program. During the recording it was possible to both to listen to the sound and to follow the phonospectrogram from the display screen, the process was completed often in less than ten minutes.

The three most common innocent murmurs, i.e. vibratory murmur, pulmonary ejection murmur and venous hum, and the five most common congenital heart defects, i.e. ventricular septal defect (VSD), aortic valve stenosis (AS), pulmonary valve stenosis (PS), patent ductus arteriosus (PDA) and atrial septal defect (ASD), are presented as examples of the whole data set.

To differentiate between innocent and pathological murmurs, 50 vibratory innocent heart murmurs, 25 innocent pulmonary or aortic ejection murmurs and 50 mild pathological systolic murmurs with intensity equal to or less than grade 3/6 were selected from a larger heart sound database for more detailed analysis. 14 of the selected pathological cases were small VSD, 5 hemodynamically loading ASD secundum, 8 PS with a pressure gradient less than 30 mmHg, 8 AS with a pressure gradient less than 30 mmHg, 3 bicuspid aortic valves with velocity less than 2.0 m/s, 1 mild CoA with a measured 15 mmHg RR difference, 1 hypertrophic cardiomyopathy (HOCM) with a septum thickness of 12 mm but with laminar aortic flow, 5 mitral leakage with or without prolapse (MI) and 3 tricuspid valve leakage (TI).

Figure 1 Innocent vibratory murmur in a 7-year-old. On auscultation, the typical early and midsystolic low-vibrating murmur is best heard at the left fourth intercostal space and the apex (S1 = First heart sound, S2 = Second heart sound, SM = Systolic Murmur). (more ...)

Figure 2 Innocent pulmonary ejection murmur in a 7-year-old. This healthy girl with a normal echocardiographic finding had a soft, midsystolic murmur that was second-degree at the most. The spectrogram shows a frequency of about 100 Hz. Auscultation area: left (more ...)

Figure 3 Venous hum in a 1.7-year-old. Auscultation revealed a systolic-diastolic second-degree murmur that was loudest in early diastole (DM = Diastolic Murmur). Venous hum is sometimes a misleadingly loud murmur. It is caused by flow in the jugular veins under (more ...)

Figure 4 Aortic stenosis in a 7-month-old. A third-degree coarse, stenosis-type murmur that does not last until the end of systole is heard in the aortic area. Echocardiography revealed aortic valve stenosis with a 45-mmHg flow gradient. The phonogram shows a (more ...)

Figure 5 Pulmonary stenosis in a 9-year-old. Haemodynamically insignificant pulmonary stenosis, with a 17–22 mmHg flow gradient on echocardiography. On auscultation a second-degree coarse, low-frequency systolic ejection murmur was heard. The phonogram (more ...)

Figure 6 Ductus arteriosus in a 2.6-year-old. On auscultation, a fourth-degree continuous murmur with throbbing pulses. Echocardiography showed a large ductus arteriosus but no other defects. The phonogram shows a consistent (systolic-diastolic) murmur that becomes (more ...)

Figure 7 Fairly large perimembranous VSD in a 7-month-old. A fairly large perimembranous VSD with normal pulmonary artery pressure (RESP = Respiration). On auscultation a fourth-degree systolic murmur was heard, and on palpation a small thrill was felt in the (more ...)

Figure 8 ASD in an 11-year-old. On auscultation a quiet systolic ejection murmur was heard in the pulmonary area. It resembled the physiological sound but was longer. The second heart sound was clearly and constantly split. The finding is well depicted on the (more ...)

The attached illustrations (Figures from 1, 2, 3, 4, 5, 6, 7, 8) show examples of the most typical murmurs in children. The upper part is a traditional phonocardiogram showing the relative amplitude of the sound. The lower part, i.e. the spectrogram, shows the sound intensity as colors on a frequency scale of 0–1000 Hz. The corresponding intensity scale in decibels is shown on the left. The duration, peak frequency and volume of the murmur were estimated based on a mean of three sequential beats. Volume was compared to the mean of the amplitudes of S1 and S2, obtained from the traditional phonocardiogram.

Table 2 Mean and standard deviation of patients' ages in the three groups

There were no significant differences in the duration of S1 and S2 between the pathological and physiological systolic murmurs (Table 3). Although S2 was constantly split in ASDs, the duration of S2 in these cases did not differ from the average in the series. By contrast, the pathological murmurs, measured in either absolute or relative terms, were longer than the innocent vibratory or ejection systolic murmurs (p < 0.001).

Table 3 Duration of S1, S2, systolic murmur and relative duration of systolic murmur

Although the pathological systolic murmurs were somewhat louder than the physiological ones, there were no significant differences between the groups (Relative Amplitude in Table 4). This confirmed that the series fulfilled our aim to select easily confusable cases. However, the pathological murmurs included both lower and higher frequencies than physiological ones (p < 0.001 for both low and high frequency limits).

Table 4 Relative amplitude and low and high frequency limits of systolic murmur

Sensitivity of 76 % and specificity of 84 % were achieved using a cut-off value of 65 % for the relative duration of systolic murmur to differentiate between pathological and physiological murmurs (Table 5 and Figure 9 – upper left). Using that cut-off, 12 (24%) pathological and 12 (16 %) physiological murmurs were misclassified.

Table 5 Capability of relative length to distinguish between murmurs: sensitivity is 76 % and specificity 84 %

Figure 9 Single criteria screening effectivity. (upper left) Sensitivity and specificity for the relative duration of the systolic murmur to differentiate between pathological and physiogical murmurs. (upper right) ROC for the relative duration of the systolic (more ...)

When the frequency of 190 Hz for the high frequency limit was used as a classification criterion, both sensitivity and specificity increased to 88% (Table 6 and Figure 9 – lower left). The number of misclassified cases decreased to 6 (12%) in the group of pathological murmurs and to 9 (12%) in the group of physiological murmurs.

Table 6 Capability of high frequency limit to distinguish between murmurs: sensitivity is 88 % and specificity 88 %

By combining these two criteria, optimal results were achieved at 80% of the relative duration and with a frequency of 200 Hz as the high frequency limit. Sensitivity approached 90 % and specificity 91 % (Table 7 and Figure 10). There were 5 (10 %) false negative findings and 7 (9 %) false positives.

Table 7 Combinated criteria: sensitivity 90 %, specificity 91 %

Figure 10 Sensitivity and specificity for the combined criteria around the optimal point (closest to 100% sensitivity and specificity) to differentiate between pathologial and physiological murmurs. (upper left) the highest frequency limit is 205 Hz and the relative (more ...)

Over 90 % of heart murmurs in children are physiological. Moreover, 75 % of them are innocent vibratory murmurs (39, 40, 41, 42). The physician needs to quickly confirm the benign nature of the heart murmur and thus to avoid misdiagnosis [25,43]. The combination of a spectrogram and a traditional phonocardiogram can be an adequate method for distinguishing innocent murmurs from pathological. When quantitative analysis is used systematically, the clinician should know what signals the system is processing and how. Visual analysis is the first step towards understanding automatic analysis, and it is a more reliable way to classify the findings as pathological or non-pathological than auscultation alone. Interpretation of the spectrogram helps to understand the hemodynamic events and the origin of heart sounds. Laminar flow will cause a harmonic wave movement in surrounding tissues, as exemplified by vibratory murmur with a peak frequency of approximately 150 Hz, which has also been reported in earlier studies [39,44,45]. Rapid flow will cause turbulence. The faster the flow velocity, the higher the sound frequencies [4,5]. These findings are also illustrated by the analysis of aortic stenosis or ventricular septal defect [6-10].

It is not simple to estimate absolute volume of heart sound. The point of maximal intensity of the murmur and the thickness of the chest vary and affects on the intensities of the sound components. In our study, the sound volume of murmur was calculated as a percentage of the volume of murmur compared to the mean volumes of S1 and S2. This may be misleading, as we can see in case 4. The volume of the aortic closure sound is decreased because of valvular stenosis, and the volume of murmur is over estimated compared to auscultation findings.

By analyzing the combined phono-spectrogram, it is possible to achieve the level of an experienced pediatric cardiologist in screening heart murmurs. In this study, by using special criteria, sensitivity of 90 % and specificity of 91% were attained. Both sensitivity and specificity were higher than using phonocradiographic or spectral analysis alone. The results were less good than those obtained in previous studies using the methods of advanced signal processing, pattern recognition and artificial networks [10,13-15]. In these previous studies, however, the series consisted of pathological cases with distinct and easily recognizable features of heart sounds and murmurs. We agree that it is easy to distinguish fast turbulent flow with a pressure gradient of over 25 mmHg, and in such cases phono-spectral analysis is reliable. We here examined the grey area, i.e. typical physiological cases compared to mildly pathological murmurs, and the most common pitfalls.

Typically, the screening method is allowed to include some false positive cases, but false negatives are unacceptable. In this study, ten (10) small muscular ventricular septal defects, three (3) cases of mild aortic stenosis, one (1) mild peripheral pulmonary stenosis, one (1) mitral valve prolapse with leakage and one tricuspid leakage fell below the 80 % duration criterion. However, each of these cases exceeded the 200 Hz frequency limit. Three (3) atrial septal defects and one (1) slight pulmonary valve stenosis included frequencies below 200 Hz, but each of them was sustained for over 80% of the systolic duration. Five (5) of the pathological murmurs were not caught with either of these two criteria. One (1) of them was a clinically insignificant tricuspid valve leakage, one (1) a very mild mitral valve leakage, one (1) a bicuspic aortic valve without stenosis and one (1) a bicuspic aortic valve with slight stenosis (pressure gradient 16 mmHg). The last and the worst missed pathological murmur was recorded from a three-year-old boy with hypertrophic cardiomyopathy (thickness of ventricular septum 12 mm). Seven out of 75 innocent murmurs exceeded one of the defined screening criteria. None of the ejection murmurs exceeded the 80 % duration limit, and 3 cases exceeded the frequency limit. All of these cases were aortic ejection murmurs with aortic velocity of 1.5–1.8 m/s. Of the vibratory murmurs, one (1) continued for over 80 % of the systolic duration, three (3) contained frequency components of over 200 Hz, but none of them exceeded both limits.

Slight tricuspid leakage is a harmless finding. Mitral valve leakage and bicuspic aortic valve should be recognized because they require endocarditis prophylaxis. The patient with mitral leakage was a fearful retarded boy, and the recorded signal was weak in amplitude, which is why the highest frequencies were missed in the analysis, but the murmur was sustained until the beginning of the second heart sound, implying a pathological finding. The murmurs due to a bicuspic aortic valve without stenosis or with slight stenosis did not differ from the hyperkinetic aortic ejection murmur. The murmur of the patient with hypertrophic cardiomyopathy resembled vibratory innocent murmur [39]. The difference between these murmurs was hardly seen in the spectrogram. Typically, the frequency of vibratory innocent murmur decreases within the systolic interval, while in this case of hypertropic cardiomyopathy it remained constant. The ECG finding was pathological, and the diagnosis was known beforehand. This particular murmur was rather loud, 31% in spectral analysis and 3/6 estimated clinically. However, 10 (20%) of vibratory murmurs were equal to or louder than 30% measured from phonogram.

As the examples show, the method did not recognize hypertrophic cardiomyopathy or mild aortic valve malformation and failed with the mitral leakage, because the recording was suboptimal. We recommend an echocardiography, when the vibratory or aortic ejection murmur is repeatedly found in adolescent age or when the recording is suboptimal. Naturally, young babies need always an echocardiography, because their pulmonary pressure may still be high and the results of heart sound analysis can be misleading.

Modern electronic stethoscopes are suitable for heart sound recording. Visual presentation of the auscultation finding provides an opportunity to study more objectively and quantitatively the timing, quality (frequency contents) and intensity of different heart sound and murmur components and helps us to understand and learn about the hemodynamics of the heart and the origin of cardiac murmurs.

Computer analysis is a powerful tool in teaching an unexperienced listener the art of auscultation [4,11,46]. Furthermore, we believe that computer analysis can be a promising new diagnostic method that has a potential to reduce medical costs by eliminating unnecessary referrals, visits to specialists and echocardiograms, especially when the systolic murmur is weak. Recording of auscultation findings also makes teleconsultations possible [47-49] and gives an opportunity to save cardiac or respiratory sounds, phonograms and spectrograms in electronic patient documents.

Phono-spectrographic analysis of heart murmur in children is a useful additional method for the clinical approach and an interesting educational tool as well. The 90 % sensitivity level of the tool may well be at the level of a trained paediatrician. However, it should be considered a single, suboptimal screening tool. As the case of HOCM indicates, ECG and echocardiography are advisable if the intensity of the murmur is fairly high and frequently audible. Patient history and careful physical examination remain the cornerstones in clinical practice.

The author(s) declare that they have no competing interests.

ALN carried out most of the clinical work, extracted the numerical data for the statistical analysis and wrote the clinical aspects of the manuscript. SL designed the recording system, performed the statistical analysis, wrote the technical description of the system, and reviewed the manuscript. AA participated in the system and study design and data acquisition. RS conceived of the study, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript.

The pre-publication history for this paper can be accessed here:

http://www.biomedcentral.com/1471-2431/7/23/prepub