Deep Learning Algorithm for Automated Cardiac Murmur Detection via a Digital Stethoscope Platform

Background Clinicians vary markedly in their ability to detect murmurs during cardiac auscultation and identify the underlying pathological features. Deep learning approaches have shown promise in medicine by transforming collected data into clinically significant information. The objective of this research is to assess the performance of a deep learning algorithm to detect murmurs and clinically significant valvular heart disease using recordings from a commercial digital stethoscope platform. Methods and Results Using >34 hours of previously acquired and annotated heart sound recordings, we trained a deep neural network to detect murmurs. To test the algorithm, we enrolled 962 patients in a clinical study and collected recordings at the 4 primary auscultation locations. Ground truth was established using patient echocardiograms and annotations by 3 expert cardiologists. Algorithm performance for detecting murmurs has sensitivity and specificity of 76.3% and 91.4%, respectively. By omitting softer murmurs, those with grade 1 intensity, sensitivity increased to 90.0%. Application of the algorithm at the appropriate anatomic auscultation location detected moderate-to-severe or greater aortic stenosis, with sensitivity of 93.2% and specificity of 86.0%, and moderate-to-severe or greater mitral regurgitation, with sensitivity of 66.2% and specificity of 94.6%. Conclusions The deep learning algorithm's ability to detect murmurs and clinically significant aortic stenosis and mitral regurgitation is comparable to expert cardiologists based on the annotated subset of our database. The findings suggest that such algorithms would have utility as front-line clinical support tools to aid clinicians in screening for cardiac murmurs caused by valvular heart disease. Registration URL: https://clinicaltrials.gov; Unique Identifier: NCT03458806.

Automated cardiac auscultation for detection of pathologic heart murmurs

Experienced cardiologists can usually recognize pathologic heart murmurs with high sensitivity and specificity, although nonspecialists with less clinical experience may have more difficulty. Harsh, pansystolic murmurs of intensity grade > or = 3 at the left upper sternal border (LUSB) are likely to be associated with pathology. In this study, we designed a system for automatically detecting systolic murmurs due to a variety of conditions and examined the correlation between relative murmur intensity and likelihood of pathology. Cardiac auscultatory examinations of 194 children and young adults were recorded, digitized, and stored along with corresponding echocardiographic diagnoses, and automated spectral analysis using continuous wavelet transforms was performed. Patients without heart disease and either no murmur or an innocent murmur (n = 95) were compared to patients with a variety of cardiac diagnoses and a pathologic systolic murmur present at the LUSB (n = 99). The sensitivity and specificity of the automated system for detecting pathologic murmurs with intensity grade > or = 2 were both 96%, and for grade > or = 3 murmurs they were 100%. Automated cardiac auscultation and interpretation may be useful as a diagnostic aid to support clinical decision making.

An Assessment of the Accuracy of Home Blood Pressure Monitors When Used in Device Owners

OBJECTIVE

To examine the accuracy of home blood pressure (BP) devices, on their owners, compared to auscultatory reference standard BP measurements.

METHODS

Eighty-five consecutive consenting subjects ≥18 years of age, who owned an oscillometric home BP device (wrist or upper-arm device), with BP levels between 80-220/50-120 mm Hg, and with arm circumferences between 25-43 cm were studied. Pregnancy and atrial fibrillation were exclusion criteria. Device measurements from each subject's home BP device were compared to simultaneous 2-observer auscultation using a mercury sphygmomanometer. Between-group mean comparisons were conducted using paired t-tests. The proportion of patients with device-to-auscultatory differences of ≥5, 10, and 15 mm Hg were tabulated and predictors of systolic and diastolic BP differences were identified using linear regression.

RESULTS

Mean age was 66.4 ± 11.0 years, mean arm circumference was 32.7 ± 3.7 cm, 54% were female and 78% had hypertension. Mean BPs were 125.7 ± 14.0/73.9 ± 10.4 mm Hg for home BP devices vs. 129.0 ± 14.7/72.9 ± 9.3 for auscultation (difference of -3.3 ± 7.3/0.9 ± 6.1; P values <0.0001 for systolic and 0.17 for diastolic). The proportion of devices with systolic or diastolic BP differences from auscultation of ≥5, 10, and 15 mm Hg was 69%, 29%, and 7%, respectively. Increasing arm circumference was a statistically significant predictor of higher systolic (parameter estimate 0.61 per cm increase; P value 0.004) and diastolic (0.38; 0.03) BP.

CONCLUSIONS

Although mean differences from 2-observer auscultation were acceptable, when tested on their owners, most home BP devices were not accurate to within 5 mm Hg. Ensuring acceptable accuracy of the device-owner pairing should be prioritized.

Application of adaptive filters to noninvasive acoustical detection of coronary occlusions before and after angioplasty

Previous studies have indicated that coronary stenoses produce sounds due to the turbulent blood flow in these vessels [1]-[10]. Measurement of these signals forms the basis of our noninvasive approach to the detection of coronary artery disease. It is during diastole that coronary blood flow is maximum and the sounds associated with turbulent blood flow through partially occluded coronary arteries would be loudest [1]-[10]. Isolated diastolic heart sounds taken from recordings made at the patient's bedside were modeled using the autoregressive (AR) and autoregressive moving average (ARMA) methods [4], [7] after adaptive line enhancement (ALE). Decisions were made in a blind fashion without prior knowledge of whether a given recording was made before or after angioplasty. Resulting model frequency spectra showed greater high-frequency components (between 400 and 800 Hz) in preangioplasty patients, and a consistent shift in amplitude of the second pole pairs of the AR and ARMA methods with surgery. Blind assessment, based on frequency spectra and poles, correctly classified the diastolic recordings in 18 of 20 cases. These results provide strong evidence supporting our hypothesis that coronary stenoses produce detectable sounds during diastole [1]-[10].

Evaluation of remote stethoscopy for pediatric telecardiology

OBJECTIVE

To investigate the interobserver reliability and diagnostic validity of a commercial electronic stethoscope for pediatric telecardiology.

MATERIALS AND METHODS

Pairs of blinded pediatric cardiologists made independent diagnoses, recommendations concerning follow-up echocardiography, and specific judgments regarding heart sounds, murmurs, and congenital heart disease using an electronic (ES) or an acoustic (AS) stethoscope on 78 pediatric cardiology outpatients and at a distance of 450 km (280 miles) with 38 telemedicine cardiology outpatients. The kappa statistic (K) indexed the instruments' interexaminer reliabilities. The validity of ES was measured by K for ES versus AS and by the percentage of cases where the findings for ES and AS differed sufficiently to suggest an important ES screening error.

RESULTS

For heart disease, AS, ES, and tele-ES reliabilities were satisfactory (K = 0.80, 0.67, and 0.80, respectively), as were AS agreement with hands-on ES (K = 0.65) and with tele-ES (K = 0.64). The AS and ES reliabilities and ES/AS agreement were also satisfactory for systolic regurgitant and diastolic pulmonic murmurs (K = 0.63-0.78) but were unsatisfactory for evaluable heart sounds and other murmurs (K = 0.16-0.60). The ES yielded clinically important disagreements with AS in 5.4% of the clinic cases and 10.5% of the telemedicine cases (P = 0.67). In determining the need for additional work-up (echocardiography) or follow-up appointments, hands-on ES and tele-ES had a combined accuracy of 92%, with a sensitivity of 88% and a specificity of 97%.

CONCLUSIONS

Hands-on ES provided reliable and valid screening for congenital heart disease. Tele-ES was highly reliable but had reduced diagnostic validity. Examiner blinding, bandwidth limitations, and artificial restrictions on the remote assistant may have contributed to this reduced performance. As these factors are easily correctable, we regard the ES as a highly promising tool for pediatric telecardiology.

Usefulness of a new sound spectral averaging technique to distinguish an innocent systolic murmur from that of aortic stenosis

We present a new method to record and display heart sounds that uses a hand-held computer and stethoscopic recording device. It allows for rapid spectral and waveform displays of murmurs and provides a means for signal averaging of spectral frequency content. Compared with aortic stenosis, innocent murmurs primarily contain frequencies of <300 Hz and persist for a shorter duration at the upper-frequency levels. This method provides for rapid characterization of innocent murmurs, a means for comparison with other murmurs, and transmission of acoustic data to distant sites.

Auscultation of the normally functioning prosthetic valve

Before prosthetic function can be adequately assessed, physicians must be acquainted with the normal acoustic events associated with these devices. The auscultatory findings of the most commonly used devices--ball, disk, porcine, and bileaflet valves--are reviewed. The mechanisms of sound production and their timing are discussed.

Heart sound segmentation of pediatric auscultations using wavelet analysis

Auscultation is widely applied in clinical activity, nonetheless sound interpretation is dependent on clinician training and experience. Heart sound features such as spatial loudness, relative amplitude, murmurs, and localization of each component may be indicative of pathology. In this study we propose a segmentation algorithm to extract heart sound components (S1 and S2) based on it's time and frequency characteristics. This algorithm takes advantage of the knowledge of the heart cycle times (systolic and diastolic periods) and of the spectral characteristics of each component, through wavelet analysis. Data collected in a clinical environment, and annotated by a clinician was used to assess algorithm's performance. Heart sound components were correctly identified in 99.5% of the annotated events. S1 and S2 detection rates were 90.9% and 93.3% respectively. The median difference between annotated and detected events was of 33.9 ms.

Single cuff comparison of two methods for indirect measurement of arterial blood pressure: standard auscultatory method versus automatic oscillometric method

With the introduction of automatic oscillometric systems for indirect measurement of arterial blood pressure (e.g. DINAMAP), the problem of correspondence between that method and the standard auscultatory method arose. For an exact and valid comparison of two methods, for physiological and methodological reasons, both measurements have to be performed simultaneously and using only one single cuff. Applying a methodological approach fulfilling these preconditions and offering in addition the advantage of a graphic documentation of the individual measurement cycles, we were able to investigate both methods in a sample of 216 comparative experiments. We found a mean difference (auscultatory method minus oscillometric method) between the methods (bias) of -0.82 mm Hg (-0.109 kPa) for systolic pressure, 1.25 mm Hg (0.166 kPa) for diastolic pressure and 1.00 mm Hg (0.133 kPa) for mean pressure. Except for systolic pressure, the differences were statistically significant (paired t-test and analysis of variance). We also obtained a significant negative correlation between the differences and the average of both pressure values (decreasing error with increasing pressure) and a significant negative correlation between differences and heart rate (decreasing differences with increasing heart rate). The conclusions drawn from the comparative study are that, although the oscillometric method exhibited a tendency to higher systolic and lower diastolic measurement values compared to the auscultatory method, both methods are well comparable and the differences are below the level of physiological and clinical significance.

Remote auscultatory patient monitoring during magnetic resonance imaging

A system for patient monitoring during magnetic resonance imaging (MRI) is described. The system is based on remote auscultation of heart sounds and respiratory sounds using specially developed pickup heads that are positioned on the precordium or at the nostrils and connected to microphones via polymer tubing. The microphones operate in a differential mode outside the strong magnetic field to reduce various sources of interference from the MRI equipment. After amplification, the signal is transmitted as infrared light to a small, battery-operated receiver and a headphone set. Thus, the patient can be simultaneously auscultated both inside and outside the shielded MRI room by infrared transmission through a metal mesh window. Bench tests of the system show that common mode acoustic noise is suppressed by approximately 30 dB in the frequency region of interest (100-1,000 Hz), and that polymer tubing having a diameter of approximately 2 mm can be used for efficient sound transmission. Recordings in situ show satisfactory detection of both heart sounds and respiratory sounds, although the signal is somewhat masked by noise during imaging. A clinical test incorporating 17 sedated or anesthetized patients was also performed. In all but four cases, the quality of the breath and heart sounds was regarded as acceptable or better.

Atrial pacing thresholds measured in anesthetized patients with the use of an esophageal stethoscope modified for pacing

Transesophageal atrial pacing (TAP) with the use of standard, thermistor-equipped, esophageal stethoscopes, modified for pacing by incorporation of a 4-French, bipolar TAP probe (pacing esophageal stethoscope [PES]), was evaluated in 100 adult patients under general anesthesia. A commercially available TAP pulse generator supplied 10-ms pulses with current variable between 0 and 40 mA. Pacing distances (in centimeters) were measured from the infraalveolar ridge to midway between PES electrodes (1.5-cm interelectrode distance). Pacing thresholds (milliamperes) were measured at the point of a maximum-amplitude P-wave (PMAX) in the bipolar esophageal electrogram and points 1 cm proximal or 1, 2, or 3 cm distal to PMAX. TAP (70-100 beats per min) was used for sinus bradycardia less than or equal to 60 beats per min (36 patients) or atrioventricular (AV) junctional rhythm (2 patients) and blood pressure changes with TAP documented. In male patients (n = 49), PMAX was 32.7 +/- 0.3 cm (mean +/- SE) and minimum pacing threshold 5.1 +/- 0.4 mA (range, 1-13 mA) at 33.6 +/- 0.3 cm (range, 30-37 cm). In female patients (n = 51), PMAX was 30.4 +/- 0.4 cm and minimum pacing threshold 4.4 +/- 0.4 mA (range, 2-14 mA) at 31.1 +/- 0.4 cm (range, 26-40 cm). TAP produced an average 13-16 mmHg increase in systolic, diastolic, or mean arterial pressure in patients with sinus bradycardia or AV junctional rhythm. There were no subjective patient complaints (epigastric discomfort, dysphagia) that could be attributed to TAP; objective evaluation (esophagoscopy) was not performed.(ABSTRACT TRUNCATED AT 250 WORDS)

Hemodynamic correlates of the third heart sound and systolic time intervals

Bedside diagnostic tools remain important in the care of patients with heart failure. Over the past two centuries, cardiac auscultation and phonocardiography have been essential in understanding cardiac pathophysiology and caring for patients with heart disease. Diastolic heart sounds (S3 and S4) and systolic time intervals have been particularly useful in this regard. Unfortunately, auscultation skills have declined considerably, and systolic time intervals have traditionally required carotid pulse tracings. Newer technology allows the automated detection of heart sounds and measurement of systolic time intervals in a simple, inexpensive, noninvasive system. Using the newer system, the authors present data on the hemodynamic correlates of the S3 and abnormal systolic time intervals. These data serve as the foundation for using the system to better understand the test characteristics and pathophysiology of the S3 and systolic time intervals, and help to define their use in improving the bedside diagnosis and management of patients with heart failure.

Evaluation of transesophageal atrial pacing stethoscope in adult surgical patients under general anesthesia

Sinus bradycardia (SB) and atrioventricular junctional rhythm (AVJR) commonly cause circulatory insufficiency in anesthetized surgical patients. Treatment is usually with drugs, which can be ineffective or have adverse effects. Cardiac pacing might be preferred, but the transvenous or epicardial routes are too invasive for routine use, and transcutaneous pacing fails to preserve atrial transport function. Transesophageal atrial pacing (TAP) lacks these disadvantages, yet unavailability of inexpensive products has prevented more widespread use. Therefore, a pacing esophageal stethoscope (PES) fabricated by addition of bipolar electrodes to disposable esophageal stethoscopes routinely used for intraoperative monitoring, was evaluated in 100 anesthetized adults. TAP thresholds (10-msec pulses) and hemodynamic effects of TAP as treatment for incidental SB (< or = 60 beats/min) or AVJR were determined. Minimum TAP thresholds (mean +/- standard error) in 48 males were 7.3 +/- 0.3 mA and in 51 females were 8.5 +/- 0.4 mA. Corresponding inferior alveolar ridge-to-electrode distances were 32.5 +/- 0.2 and 30.4 +/- 0.2 cm. For 48 patients with SB < or = 60 beats/min (54 +/- 1 beats/min), TAP (81 +/- 1 ppm) produced average 15, 11, and 14 mmHg increases in systolic, diastolic, and mean arterial pressure, respectively (P < 0.001). For 11 patients with AVJR (71 +/- 5 beats/min), TAP (92 +/- 3 ppm) produced average 23 and 15 mmHg increases in systolic and mean arterial pressure, respectively (P < 0.05). There were no apparent complications of TAP. TAP with a PES appears practical, safe, and effective for prophylaxis and treatment of SB or AVJR in anesthetized surgical patients.

Effect of teaching and type of stethoscope on cardiac auscultatory performance

BACKGROUND

Auscultation of the heart is a routine procedure. It is not known whether auscultatory skills can be improved by teaching or with the use of an advanced stethoscope.

METHODS

This study was a randomized trial with a 2 x 2 factorial design. Seventy-two house officers were randomized to a simple or an advanced stethoscope and to a 4-hour course in auscultation or no course. The doctors auscultated 20 patients' hearts and categorized findings as normal or as one or more of 5 categories of heart diseases. Patients were selected such that 16 had a known heart disease as well as a corresponding murmur and 4 had no heart disease or murmur. Auscultatory performance was assessed as concordance with echocardiographic findings and interobserver variation.

RESULTS

Doctors using the advanced stethoscope diagnosed 35% of the patients correctly, as compared with doctors using the simple stethoscope who did 33% of the patients (P = .27). Similarly, 34% of the patients were diagnosed correctly by doctors who had received teaching as compared with 33% of those who were by doctors who had received no teaching (P = .41). The kappa values were higher for doctors who had received teaching for aortic stenosis (0.43 vs 0.28, P = .004) and ventricular septum defect (0.07 vs 0.01, P = .003). There was no difference between groups for any other single murmur or for the detection of murmurs as such.

CONCLUSION

Heart auscultation findings were in poor accordance with echocardiographic findings and had high interobserver variation. Neither outcome improved to any important extent with the subjects' use of an advanced stethoscope or attending of a course in heart auscultation.

Diagnosis and treatment of cardiac arrhythmias

Cardiac arrhythmias are probably more common in horses than in any other domestic animal species. The most frequent clinical complaint associated with cardiac arrhythmias is exercise intolerance. Physical examination is characterized by auscultation abnormalities such as fast or slow heart rate, irregular rhythm, extra sounds, long pauses, or abnormal heart sounds. The electrocardiogram is used to make a definitive diagnosis of the dysrhythmia. Other laboratory and cardiac function tests are employed to determine the etiology and to assess the significance of the arrhythmia. Antiarrhythmic therapy is given when clinical signs specifically related to the arrhythmia are present, when hemodynamic parameters are compromised by the arrhythmia, or when the ECG reveals abnormalities that put the patient at risk for development of more severe arrhythmias. The cardiovascular drugs most frequently used are digoxin and quinidine. Digoxin is most commonly used for supraventricular arrhythmias, especially arrhythmias characterized by fast heart rates. Quinidine is very effective for short-term treatment of ventricular and supraventricular arrhythmias but must be used with caution because of the potential for toxic side effects. The cardiac arrhythmias due to vagal tone (sinus arrhythmia, sinus block, sinus arrest, sinus bradycardia, wandering pacemaker, first-degree AV block, and second-degree AV block) that are found in resting horses are generally considered to be normal and generally do not require therapy.

Optimization of Cardiac Resynchronization Devices Using Acoustic Cardiography: A Comparison to Echocardiography

Optimization of pacemaker settings for cardiac resynchronization therapy (CRT) remains challenging and problematic. Several noninvasive methods are offered to customize the programmed parameters for individual patients, but so far only echocardiographic imaging has established itself as an accepted method. The authors examined the value of acoustic cardiography as a fast and more cost-efficient alternative to established echocardiographic imaging techniques for the optimization of CRT devices. The atrioventricular delay in 22 subjects with implanted CRT devices was independently optimized using echocardiography (Doppler transmitral flow) as well as acoustic cardiography, and the recommended settings from each method were later compared. Doppler echocardiography and acoustic cardiography recommendations matched within a mean value +/- SD of 17+/-16 milliseconds and gave a correlation coefficient of r=0.90 (p<0.001). In 17 of the 22 cases (77.3%), the difference between echocardiographic and acoustic cardiogram CRT optimization results was <or=20 milliseconds. Furthermore, the echocardiographic transmitral flow pattern was not significantly different for the setting independently chosen by the echocardiographic expert and the acoustic cardiographer for the cases with a difference of >20 milliseconds (22.7%). In addition, it took less time for the acoustic cardiogram to collect sufficient information to make a recommendation, and it was found that the acoustic cardiogram data trend is easier to interpret.

Observer reliability in detecting surreptitious random occlusions of the monaural esophageal stethoscope

The esophageal stethoscope is used often during anesthesia to monitor ventilation and cardiac function. Deficiencies in observer vigilance may limit the effectiveness of this monitoring instrument. The aim of this study was to determine how long it took for an observer to detect a surreptitiously occluded monaural esophageal stethoscope in the setting of clinical anesthesia. During routine anesthesia, where an esophageal stethoscope was in use, a computer-guided device would artificially, silently, and at random time intervals, occlude the stethoscope tubing. Personnel using the stethoscope noted when they perceived the absence of stethoscope sounds. We studied 320 stethoscope occlusions in 32 patients. The time between stethoscope occlusion and detection was 34 +/- 59 seconds (mean +/- SD). Eighty-seven percent of detections were made in less than 60 seconds. However, 13% of detections were delayed for more than 60 seconds, and 2.3% for more than 240 seconds. While anesthesia personnel using an esophageal stethoscope could detect most stethoscope occlusions, failure to appreciate such episodes occurred in a small but significant number of cases. This suggests that the esophageal stethoscope has some definite limitations as a continuous monitor and that other monitoring techniques, such as oximetry, capnography, and ventilator disconnect alarms, as well as visual/tactile inspection of the patient, should be used as well.

An electronic stethoscope is judged better than conventional stethoscopes for anesthesia monitoring

A prototype electronic monitoring stethoscope was constructed from readily available, high-quality components. It consisted of a conventional precordial or esophageal probe connected to a microphone by a rubber adapter. The microphone was connected by lightweight wire to an amplifier and headphones. Twenty-one anesthesia clinicians evaluated the stethoscope and responded to a multiple-choice preference questionnaire. The electronic stethoscope was judged to perform better than the conventional stethoscope in most categories evaluated. The electronic device was perceived to be louder, clearer in sound reproduction, more efficacious for monitoring, and easier to use continuously, and its headphones were considered more comfortable than the conventional earpiece. Based on our results, we conclude that amplified stethoscopes have the potential to improve monitoring. Further development of electronic stethoscope monitoring seems warranted and is continuing.

Intermittent auscultation of the fetal heart rate: a review of general principles

Multiple randomized clinical trials have been unsuccessful in providing evidence of efficacy of electronic fetal monitoring; thus, there is renewed interest in intermittent auscultation during labor for women with low-risk pregnancies. Auscultation must be used with palpation or external or internal electronic monitoring of uterine contractions. Auscultation and palpation require education, experience, and competency validation at regular intervals. Institutional policies and standards of care are mandatory for intermittent auscultation. Concerns exist regarding the personnel costs for auscultation; however, these costs may ultimately be shown to be offset by significant benefits in improved outcomes and patient satisfaction.