Skip to main content
CHEST Pulmonary logoLink to CHEST Pulmonary
. 2024 Jun 22;2(4):100079. doi: 10.1016/j.chpulm.2024.100079

Machines Are Learning Chest Auscultation. Will They Also Become Our Teachers?

Hans Pasterkamp a,∗, Hasse Melbye b
PMCID: PMC13418314  PMID: 42548356

Abstract

Great strides in the development of machine learning techniques are bringing applications of artificial intelligence to ever more areas of clinical medicine. Their potential in the evaluation of visual images and in speech recognition is well established. Recently, the capabilities of machine hearing have been also applied to chest auscultation (ie, the automated analysis, characterization, and classification of heart and lung sounds). Comparing strengths and limitations of human vs machine hearing can help to put these developments in perspective. Humans have multisensory perception (ie, they receive visual and tactile information while auscultating). Humans also surpass machines in the ability to focus attention on listening for specific sounds in noisy environments. Together with information on a patient’s history and presumed medical diagnosis, and with frequent repetition, chest auscultation remains a trainable and valuable human skill. Advantages of machine hearing of chest sounds with digital stethoscopes include not only objective acoustic analysis but also storage of data that allows comparisons over time, presentation in audiovisual format, and wireless communication. Machines can support patient management by relating acoustic analyses to clinical diagnoses, serving as decision support for further investigations, and by monitoring of patients over time. The potential of machines to become teachers of chest auscultation is only now coming into focus. In the near future, assessment of chest sounds will largely remain in the domain of traditional acoustic stethoscopes. However, machines may well be used for training students in different health care professions and nonmedical caregivers, provided that humans remain part of the process.

Key Words: artificial intelligence; education, medical; heart auscultation; machine learning; respiratory sounds


Artificial intelligence (AI) (ie, computer systems that take on tasks which previously had been limited to humans) has found increasingly important applications in public health, image analysis, retrieval of medical information, clinical trial performance, and organizational optimization.1 Machine learning (ML), as an application of AI, trains machines with known data to generate algorithms that predict properties of new and unknown data. Automated recognition of complex acoustical data by ML (eg, responding to human voice commands) has already become widely accessible and is used in our daily lives.

In clinical medicine, the characterization of respiratory sounds that can be recorded without stethoscope (eg, cough, snoring) by smart devices based on ML is translating into new applications and diagnostic opportunities.2 The assessment of heart and lung sounds is still being performed using traditional acoustic stethoscopes for chest auscultation, but the benefits of ML in digital smart stethoscopes have raised attention in recent years, and several devices have already become commercially available. To date, the clinical uses of smart stethoscopes has been mostly in cardiovascular and pediatric medicine. Their use in assisting educators who teach chest auscultation to students and trainees in the health care professions remains limited, but we are close to a point where ML is moving into a historic domain of humans, in this case that of teachers of chest auscultation. Given the decline of physical examination skills over recent decades and the related decreasing confidence of clinicians to teach these skills,3 the potential of ML to assume this role is growing.

The objective of our report is to review differences and similarities between human hearing and the processing of chest sounds by ML. We compare the strengths and limitations of human auditory perception with those of ML and present a view of machines as teachers of chest auscultation where humans remain essential participants in future technologic advances.

Study Design and Methods

We used PubMed advanced searches on (auscultation) AND ((lung) OR (heart)) AND ((machine learning) OR (artificial intelligence)), and on (auscultation) AND ((lung) OR (heart)) AND ((teaching) OR (education)). In addition, we also searched Google Scholar for the most recent publications on the following: machine hearing OR machine listening OR computer audition. For our narrative review, we focused on full-length publications in peer-reviewed journals and in English language, with attention to recent relevant reviews.

Evidence Review

Humans and ML

In 1819, Laennec introduced his monaural stethoscope, the first instrument for indirect auscultation (ie, aided by an intervening medium). He also offered a detailed description of respiratory sounds and their relation to lung health and disease, which was published in an English translation by John Forbes in 1829. Under the direction of the distinguished cardiologist Austin Flint, an international work group presented a uniform nomenclature of auscultatory sounds in the diagnosis of diseases of the chest in 1884.4 Respiratory sounds were classified as (1) varieties of breath sounds (nine characterizations), (2) adventitious sounds (nine types), and (3) varieties of voiced sounds (five types).

Medical trainees learned about chest sounds from written descriptions, verbal instructions by their mentors at the bedside, and continuing self-supervised learning in clinical practice. In 1924, the stethophone, an electric stethoscope developed by Bell Laboratories, offered the opportunity to teach about lung sounds with multiple simultaneous listening units.5 Clinical assessment of heart sounds and murmurs was ahead of lung sounds by directing a focus on the limitations of human hearing.6 The repetitive acoustic events of the heart and their relation to mechanisms of sound generation led to the development of phonocardiography.7 Selective attention to the audible components of the cardiac cycle was recognized as a most difficult skill. Using visual display of ECG and sounds, methods were developed to allow students and instructors to silence selected parts of recordings and thereby focus their attention.8

Adapting the nomenclature for respiratory sounds to new understandings of respiratory diseases based on spirometry and radiography, a simplified categorization of adventitious sounds, such as rales (crackling or bubbling) and rhonchi (wheezes), was proposed by an ACCP-ATS Joint Committee in 1975.9 Ambiguity of the term rhonchus as it was used originally by Laennec led to a change of terms from rales and rhonchi to crackles (fine and coarse) and wheezes (high and low pitched).10 The more recent classification by a task force of the European Respiratory Society kept the term rhonchus for low pitched wheezes.11

The introduction of sound spectrography12 and time-expanded analysis of respiratory sound waves13 paved the way for further developments of stethoscopes. Electronic stethoscopes enhance the human hearing of chest sounds by amplification, filtering, and reduction of extraneous noise, whereas the digital format of recorded chest sounds takes this further by processing for storage, wireless transmission, and computer analysis.14,15 First reports of neural network analyses of heart sounds16 and lung sounds17 appeared in the 1990s. The most recent developments apply various ML techniques to the automated characterization of heart and lung sounds (Fig 1).18

Figure 1.

Figure 1

Advances in stethoscopes and methods for the recording and teaching of chest sounds have accelerated in recent decades. AI = artificial intelligence.

For the learning of lung sounds, advances included a progression from individual or group teaching based on recorded audio tapes and CDs to multimedia presentations that included the graphic display of their relation to breath phases and airflow19 and of their sound characteristics.20 Computer-based learning of chest auscultation was recognized as a potential substitute for conventional teaching methods,21 and auscultatory simulation of chest sounds with manikins has seen increased use over the past 2 decades.22

The effectiveness of teaching cardiac auscultation can be increased with teaching methods that emphasize repetition and normal-abnormal comparisons of sounds, using computer-assisted and online resources.23 An added value by manikins to a low-fidelity simulation with tapes and audiovisual CDs to teach cardiac auscultation has been questioned,24 but hybrid simulation with an AI-supported and ECG-based virtual pathology stethoscope25 may become a more convincing step forward. Also, augmented reality auscultation may offer to reliably synchronize educational respiratory sounds with the breathing of a simulated patient, using breath measurement methods that are integrated in the stethoscope chest piece.26

The potential of machines that learn to identify human sounds is illustrated by the progress from Audrey, a device developed by Bell Laboratories that could identify 10 spoken numbers in 1955, to the present capability of machine hearing (eg, to confirm student identity by recognition of their voices in distance education).27 In the 1990s, first reports appeared on the use of neural networks in the analysis of heart sounds28 and lung sounds17 as a foundation of a coming era of new auscultation, given the rapid progress in the development of deep learning machines.29

Classical ML learning uses training sets of data that have been labeled for supervised learning or used without labels for unsupervised learning.30 In supervised learning, algorithms are developed to process new data to best fit the desired classification as the output (eg, normal, crackles, or wheezes in the characterization of respiratory sounds). In unsupervised learning, the data are processed according to algorithms that sort features into clusters. Further processing, typically based on sounds represented as spectrograms (Fig 2), determines the best distinct separation of clusters which then are compared with a classification by humans of at least a representative sample of the data.18 By comparing results with other than human auscultatory descriptions, or meta-information (eg, patient demographics, specific medical diagnoses,31 anatomic sites of recording), more complex acoustic presentations of respiratory disease can potentially be identified with no simple correlate in the human classification of adventitious sounds.32

Figure 2.

Figure 2

Respiratory sounds during three respiratory cycles, recorded from a male patient with COPD. Computer analysis presents a mel spectrogram with frequencies (Hz) on a logarithmic scale, similar to the decibel (dB) scale of loudness. Arrows point to crackles in early inspiration, wheezes in first and third expiration, and two artifactual clicks in second expiration. a = artifactual clicks; C = crackles; Exp. = expiration; Insp. = inspiration; W = wheezes.

The development of artificial neural networks (ANNs) has been inspired by the biological networks of the human brain. These ANNs are constructed with nodes that are somewhat akin to groups of neurons in biological networks. These nodes are arranged in an input layer, feeding into one or more subsequent hidden layers that feed into an output layer. ANNs allow machines to use algorithms that can fine tune the connection between nodes by adjusting their relative contribution. This adjustment of weights reduces prediction errors and can increase the accuracy of the model output. Deep learning refers to the use of multiple hidden layers in ANNs,30 a technique that has dramatically improved ML in speech recognition.33

In human hearing, the external and middle ear condition incoming acoustic signals by damping or increasing sounds. Next, the cochlea transforms the analog sound pressure wave according to their frequencies and temporal appearance into electrical nerve impulses. These digital images are stored in the cochlear nucleus as ultra-short-term or echoic memory for just a few seconds, which allows to focus attention on what was just heard. This decides the likelihood of the information being moved up to short-term acoustic memory in the hippocampus and further up to the auditory cortex. Learned acoustic patterns can be recalled and shared by speaking or in writing.

In machine hearing, the initial steps of signal conditioning and subsequent passage through a filter bank (eg, Mel and/or gammatone filters that transform complex sounds similar to output of the cochlea to the acoustic nerve) are comparable to the pathway of human hearing (Fig 3). Training a real-time neural network that captures the nonlinearity of the inner ear based on a cochlear model of transmission appears as a promising approach to greatly increase the speed of machine hearing.34 Short-term temporary storage in a circular buffer before further steps (eg, by long short-term autoencoding that excludes or forgets parts of the information before neural processing to the final output) are not directly comparable to human hearing.35 Passing erroneous results back through the hidden layers to the input (ie, backpropagation) and reprocessing with automatically modified weights of nodes in the network has become a widely used and successful technique in ML33,36 that also has no direct correlate in human hearing. Machines can present learned acoustic patterns by audiovisual feedback that can be shared wirelessly.

Figure 3.

Figure 3

Machine learning of auscultation shares several steps in the recognition of chest sounds that are similar to human hearing (see text for details). LSTM = long short-term autoencoding.

Strengths and Limitations

Although backpropagation and self-adjustment of steps in convoluted neural networks works well for machines, the existence of such process and its potential value in the biological neural network of humans remains unclear.36 Comparing the strengths of human hearing in chest auscultation to that of machines, the multisensory perception by humans, their knowledge and focused attention, and awareness of the clinical and environmental context of the examination need to be viewed in more detail. For machine hearing, the objective quantification and documentation of chest sounds, the permanent storage of analyses, and the possibility to compare findings over space and time need to be described.

When auscultating a patient’s chest, a human examiner perceives not only the auditory input from heart and lung sounds, but processes this in context with simultaneous visual input about the site of listening and the effort and phases of the patient’s breathing. Additional input from the tactile sense allows a human examiner to process chest sounds in relation to extraneous noises from the patient’s movements, from varying contact of the stethoscope head and from the applied pressure when listening through a layer of light clothing.37

Another strength of human hearing is that of selective and divided attention, and the ability to switch between the two. In selective attention, a listener focuses on one sound at the expense of ignoring others, whereas in divided attention, multiple auditory stimuli are simultaneously processed. This is known as the cocktail party phenomenon, referring to the ability to follow conversations of interest even in noisy environments with different simultaneous speech signals. The phenomenon is also known as the cocktail party problem because decades of investigations of this complex perceptual task by researchers in the fields of physiology, neurobiology, psychophysiology, cognitive psychology, and biophysics have not delivered a complete understanding of this phenomenon.38 Computer scientists and engineers who work on machine hearing have a particular interest in simplifying this task as a learning problem, also by exploring the use of crossmodal (ie, auditory plus visual) computational models.39

Listening to chest sounds while aware of extraneous noises and knowing the patient’s history and presumed medical diagnosis creates the context in which heart and lung sounds are perceived and processed. Binaural hearing and spatial recognition of sounds is relevant for automated speech recognition in noisy environments. Auscultated chest sounds do not have a component of spatial recognition because the binaural input arises from a single stethoscope head. Although the cocktail party effect is reduced by monophonic hearing during auscultation, efferent pathways from auditory cortex to cochlea may also contribute to selective attention and unmasking of auditory stimuli in background noise.40

Strengths of machine hearing in chest auscultation are the objective quantification and characterization of heart and lung sounds, and the storage of recorded sounds and concurrent signals for further analyses and comparisons. The bandwidth of sound frequencies that can be registered by machines exceeds the general and individual limits of human auditory perception, but this is only a theoretical strength at present because the frequency range of sound transmission through the human chest falls overwhelmingly within the range of human hearing.41

Deep learning techniques for the machine hearing of chest sounds use automatic feature extraction for a visual representation of time-frequency maps.18,42 Musical sounds of wheezing can thus be quantified by their pitch and relative duration during respiratory cycles, whereas crackles can also be automatically classified as fine or coarse by their timing and frequency content.43 For heart sounds, concurrent recording of an ECG to relate physiologic events of the cardiac cycle is easier than the recording of respiratory airflow or chest movements to relate lung sounds to the respiratory cycle and depth of breathing. However, extracting breath phase information from the respiratory sound itself appears promising44,45 to avoid additional instrumentation (eg, thoracic impedance measurements to identify breath phases that were used in a bimodal repository of lung sounds).46 Such storage and sharing of recorded lung sounds in clouds of electronic data has greatly accelerated the development of machine hearing. For individual patients, the analysis and local storage of their chest sounds on digital stethoscopes offers longitudinal comparisons.47

Comparing such auscultated chest findings over time is more difficult for humans. Echoic memory permits the comparison of sequentially auscultated lung sounds at separate locations, or at one location but different depths of breathing, yet only within a few seconds. The appearance of new findings (eg, inspiratory crackles that were not heard a day prior at the same site) are not a challenge, but changes in basic sound characteristics (eg, their relation to the depth of breathing) unveil the relative weakness of human auditory memory compared with that for visual or tactile stimuli. Although these sensations are remembered equally well for the first 4 s, the accuracy for remembering auditory stimuli drops off well below visual and tactile recall after a retention interval of 30 s.48

Humans perceive the loudness of sounds not equally well within the frequency range of chest sounds, but this general weakness can be reduced by filtering and amplification with electronic stethoscopes. However, there are individual limitations due to the biological function of the middle ear in healthy humans that can explain the variable onset and degree of presbycusis in older patients.49 Of greater relevance is the effect of auditory masking in human hearing (eg, when an increased loudness of lung sounds with deeper breaths can mask the presence of fine crackles).50 Hearing clinically relevant differences between very brief sounds (eg, a benign split first heart sound vs a click from bicuspid aortic valve disease) is a challenge even to expert cardiologists and independent of their years of experience.51 In general, however, many hundredfold repetitions of attentive listening over time improve the accuracy of cardiologists’ auscultation skills.23

Potential for Machines as Teachers

Publications on learning to auscultate heart and lung sounds in the medical professions, and publications on ML of these sounds, have greatly increased in recent years, driven by advances in stethoscope technology and the application of AI techniques in acoustic analyses (Fig 4).

Figure 4.

Figure 4

Publications on machine learning and human education related to chest sounds are increasing, with the focus on lung sounds now approaching that of cardiovascular sounds.

Industry is predicting an annual growth rate for the stethoscope market of close to 6% between 2023 and 2030. Customers for stethoscopes are home health care organizations, hospitals, clinics, nurse practitioners, and paramedics/first responders. Currently, about 75% of the devices are traditional acoustic stethoscopes, but smart stethoscopes are expected to grow at around 7% annually.52 Most of their present uses are for cardiovascular53,54 and pediatric assessments,55, 56 with few for the purpose of teaching. This may change more rapidly when a growing contribution by nonmedical users (eg, parents and caregivers of young children in their home environment57) is included.

Educational needs for learning chest auscultation depend on the users in different health care professions (Table 1). A human teacher has support from teaching assistance (eg, by selecting audiovisual materials, dedicated types of manikins, and most recently also from machines that are the basis for smart stethoscopes). The latter can present auscultation findings at the bedside, using audio playback with graphic displays of waveforms and spectrograms. This can improve the classification of adventitious lung sounds58 and offer auditory training and repetitive listening to build the skill of cardiac auscultation.23

Table 1.

Educational Requirements Depend on the Different Categories of Stethoscope Users Who Can Contribute to the Adaptation and Modification of Machine Learning Teaching Modules

Stethoscope Users Focus Contribution by Users
Medical specialists and general medicine practitioners Supportive findings for differential diagnosis Providing investigation results and other meta-data
Nurses and allied health professionals Relevant changes of chest sounds over time Reporting changes after therapeutic interventions
Parents and other nonmedical caregivers Recognition of respiratory sounds Supplying lay terminology and visual observations

Teaching a standardized description of auscultatory findings remains a challenge, particularly regarding lung sounds. Recommended nomenclature, when simplified to crackles and wheezes as the two main categories of adventitious sounds, results in reasonably good observer agreement. However, disagreement increases on further characterization.59 This can reduce the clinical value of rhonchi vs wheezes as indicators of different causes of airway obstruction,60 or of fine crackles and their implications in screening for interstitial lung diseases.61,62 Other terms (eg, squawk for a brief inspiratory wheeze that is often preceded by a crackle) are unknown to many physicians,11 although of potential relevance because of their relation to interstitial and fibrotic lung diseases.63

Although “the sound repertoire of a wet sponge such as the lung is limited,”64 the automated detection and classification of chest sounds is currently less competent than the capabilities of ML to recognize and interpret other comparatively complex sounds (eg, those of human speech). Automated voice recognition has become ubiquitous in our environment because it has reached close to human performance, at least in quiet environments. Training machines to classify chest sounds based on the classification by physicians, even experts, who may disagree on the presence of heart murmurs65 or rhonchi vs wheezes,11 is clearly one limiting factor.

Deep learning techniques that are remarkably successful in automated speech recognition are also being applied to the identification of chest sounds. However, the benefit of immediate feedback from user to the machine (ie, human-in-the-loop approach66) has not been as widely implemented as it has for digital voice assistants that learn by asking for confirmation or clarification of spoken requests. Using this approach along with metadata, for example from multimodal input including video capture of respiratory rate and effort67 and application pressure of the stethoscope,68 also from demographic and diagnostic patient data, or visual display options (eg, phonocardiography,69 respirosonography19), machines could automatically adjust teaching modules to meet different educational needs. Comparison of ambient respiratory sound, audible without stethoscope, with chest sounds could be taught to separate stridor from inspiratory wheezes. Further adaptations could be made according to descriptions that may be common in specific target groups (eg, among parents70 who describe snore-like rhonchi in their young children as ruttles71), or by taking into account the language backgrounds, experience, and skill levels of medical students and physicians.72

The need for respiratory assessment on a large scale and at a distance during the recent Sars-CoV-2 pandemic has accelerated ML of respiratory sounds.2 Although the performance of digital stethoscopes to support the diagnosis of pediatric pneumonia has not met expectations so far,73 the evaluation of heart and lung sounds will continue to require direct auscultation by skilled examiners.74 An erosion of basic clinical skills due to increasing reliance on digital technology is of concern,75 but regarding chest auscultation we may soon reach a turning point where machines that have learned enough about heart and lung sounds, along with access to meta-information, might overcome the limitations of our Achilles’ ears48 and become our best teachers.

Summary

Auscultation of the chest no longer holds the respect and expectation as a preeminent diagnostic tool for cardiorespiratory disease that it held during the first century after the invention of the stethoscope. Since then, major advances in the functional assessment and imaging of heart and lungs have relegated its in-hospital role largely to screening and monitoring. However, chest auscultation remains a valuable clinical skill for effective medical triage and clinical follow-up. The application of AI in smart stethoscopes and advances in training the listening machines may indeed promote them from our overlooked assistants to our revered teachers of chest auscultation.

Funding/Support

The authors have reported to CHEST Pulmonary that no funding was received for this study.

Financial/Nonfinancial Disclosures

The authors have reported to CHEST Pulmonary the following: H. M. has received remuneration for consultancy from Medsensio AS, Tromsø, Norway. None declared (H. P.).

Acknowledgments

Author contributions: Hans Pasterkamp and Hasse Melbye contributed equally to the study design, data analysis and interpretation, and the writing of the manuscript.

Other contributions: We thank Professor George R. Wodicka, Weldon School of Biomedical Engineering at Purdue University, and Professor Lars Ailo Bongo, Department of Computer Science at the Arctic University of Norway, Tromsø, Norway, for their helpful suggestions.

Footnotes

H. Pasterkamp and H. Melbye contributed equally to this manuscript.

References

  • 1.Haug C.J., Drazen J.M. Artificial intelligence and machine learning in clinical medicine, 2023. N Engl J Med. 2023;388(13):1201–1208. doi: 10.1056/NEJMra2302038. [DOI] [PubMed] [Google Scholar]
  • 2.Kraman S.S., Pasterkamp H., Wodicka G.R. Smart devices are poised to revolutionize the usefulness of respiratory sounds. Chest. 2023;163(6):1519–1528. doi: 10.1016/j.chest.2023.01.024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Faustinella P., Jacobs R.J. The decline of clinical skills: a challenge for medical schools. Int J Med Educ. 2018;9:195. doi: 10.5116/ijme.5b3f.9fb3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Correspondent Report from the International Medical Congress in Copenhagen. Lancet. 1884;124:383–384. [Google Scholar]
  • 5.Frederick H., Dodge H. “The stethophone,” an electrical stethoscope. The Bell System Technical Journal. 1924;3(4):531–549. [Google Scholar]
  • 6.Rappaport M.B., Sprague H.B. Physiologic and physical laws that govern auscultation, and their clinical application: The acoustic stethoscope and the electrical amplifying stethoscope and stethograph. Am Heart J. 1941;21(3):257–318. [Google Scholar]
  • 7.Campbell F., Sloan A., Andrew A. An electronic phonocardiograph employing a double-beam cathode-ray oscillograph as the recording device. Br Heart J. 1952;14(2):271–275. doi: 10.1136/hrt.14.2.271. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Adolph R.J., Campbell D.J. Teaching selective attention to the cardiac cycle: The Cardio-gater. Am Heart J. 1971;82(2):215–221. doi: 10.1016/0002-8703(71)90268-7. [DOI] [PubMed] [Google Scholar]
  • 9.Burrows B. Pulmonary terms and symbols: A report of the ACCP-ATS joint committee on pulmonary nomenclature. Chest. 1975;67(5):583–593. [PubMed] [Google Scholar]
  • 10.Mikami R., Murao M., Cugell D.W., Chrétien J., Cole P., Meier-Sydow J., et al. International symposium on lung sounds: synopsis of proceedings. Chest. 1987;92(2):342–345. doi: 10.1378/chest.92.2.342. [DOI] [PubMed] [Google Scholar]
  • 11.Pasterkamp H., Brand P.L., Everard M., Garcia-Marcos L., Melbye H., Priftis K.N. Towards the standardisation of lung sound nomenclature. Eur Resp J. 2016;47(3):724–732. doi: 10.1183/13993003.01132-2015. [DOI] [PubMed] [Google Scholar]
  • 12.McKusick V.A., Jenkins J.T., Webb G.N. The acoustic basis of the chest examination; studies by means of sound spectrography. Am Rev Tuberc. 1955;72(1):12–34. doi: 10.1164/artpd.1955.72.1.12. [DOI] [PubMed] [Google Scholar]
  • 13.Murphy Jr R.L., Holford S.K., Knowler W.C. Visual lung-sound characterization by time-expanded wave-form analysis. N Engl J Med. 1977;296(17):968–971. doi: 10.1056/NEJM197704282961704. [DOI] [PubMed] [Google Scholar]
  • 14.Kim Y., Hyon Y., Woo S.-D., Lee S., Lee S.-I., Ha T., et al. Evolution of the stethoscope: Advances with the adoption of machine learning and development of wearable devices. Tuberc Respir Dis (Seoul). 2023;86(4):251. doi: 10.4046/trd.2023.0065. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Pasterkamp H., Sanchez I., Oh Y. Teleauscultation: Bringing you to the bedside. Can Respir J. 1995;2(4):237–240. [Google Scholar]
  • 16.Barschdorff D., Ester S., Dorsel T., Most E. IEEE Proceedings Computers in Cardiology. 1990. Neural network based multi sensor heart sound analysis; pp. 303–306. (23-26 September 1990) [Google Scholar]
  • 17.Rietveld S., Oud M., Dooijes E.H. Classification of asthmatic breath sounds: preliminary results of the classifying capacity of human examiners versus artificial neural networks. Comput Biomed Res. 1999;32(5):440–448. doi: 10.1006/cbmr.1999.1522. [DOI] [PubMed] [Google Scholar]
  • 18.Nguyen T., Pernkopf F. In: State of the Art in Neural Networks and Their Applications. El-Baz A.S., editor. 2023. Computational lung sound classification: a review; pp. 193–215. [Google Scholar]
  • 19.Pasterkamp H., Carson C., Oh Y. Digital respirosonography: new images of lung sounds. Chest. 1989;96(6):1405–1412. doi: 10.1378/chest.96.6.1405. [DOI] [PubMed] [Google Scholar]
  • 20.Sestini P., Renzoni E., Rossi M., Beltrami V., Vagliasindi M. Multimedia presentation of lung sounds as a learning aid for medical students. Eur Respir Journal. 1995;8(5):783–788. [PubMed] [Google Scholar]
  • 21.Mangione S., Nieman L.Z., Gracely E. Comparison of computer-based learning and seminar teaching of pulmonary auscultation to first-year medical students. Acad Med. 1992;67(10):S63–S65. doi: 10.1097/00001888-199210000-00041. [DOI] [PubMed] [Google Scholar]
  • 22.Ward J.J., Wattier B.A. Technology for enhancing chest auscultation in clinical simulation. Respir Care. 2011;56(6):834–845. doi: 10.4187/respcare.01072. [DOI] [PubMed] [Google Scholar]
  • 23.Barrett M.J., Mackie A.S., Finley J.P. Cardiac auscultation in the modern era. Cardiol Rev. 2017;25(5):205–210. doi: 10.1097/CRD.0000000000000145. [DOI] [PubMed] [Google Scholar]
  • 24.Osborne C., Brown C.W., Mostafa A. Effectiveness of high and low-fidelity simulation-based medical education in teaching cardiac auscultation: a systematic review and meta-analysis. International Journal of Healthcare Simulation. 2022;1(3):75–84. [Google Scholar]
  • 25.Yhdego H., Kidane N., Mckenzie F., Audette M. Development of deep-learning models for a hybrid simulation of auscultation training on standard patients using an ECG-based virtual pathology stethoscope. SIMULATION. 2023;99(9):903–915. [Google Scholar]
  • 26.Kono Y., Miura K., Kasai H., Ito S., Asahina M., Tanabe M., et al. Breath measurement method for synchronized reproduction of biological tones in an augmented reality auscultation training system. Sensors. 2024;24(5):1626. doi: 10.3390/s24051626. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Portugal D., Faria J.N., Belk M., Martins P., Constantinides A., Pietron A., et al. Continuous user identification in distance learning: a recent technology perspective. Smart Learning Environments. 2023;10(1):38. [Google Scholar]
  • 28.Durand L.-G., Pibarot P. Digital signal processing of the phonocardiogram: review of the most recent advancements. Crit Rev Biomed Eng. 1995;23(3-4):163–219. doi: 10.1615/critrevbiomedeng.v23.i3-4.10. [DOI] [PubMed] [Google Scholar]
  • 29.Kim Y., Hyon Y., Lee S., Woo S.-D., Ha T., Chung C. The coming era of a new auscultation system for analyzing respiratory sounds. BMC Pulm Med. 2022;22(1):119. doi: 10.1186/s12890-022-01896-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Smith S. In: Skoe E., Parker A., editors. Vol. 43. Thieme Medical Publishers, Inc; 2022. Translational applications of machine learning in auditory electrophysiology; pp. 240–250. (Seminars in Hearing). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Matek C. More than just sound: Harnessing metadata to improve neural network classifiers for medical auscultation. Patterns. 2022;3(1):1–2. doi: 10.1016/j.patter.2021.100426. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Heitmann J., Glangetas A., Doenz J., Dervaux J., Shama D.M., Garcia D.H., et al. DeepBreath—automated detection of respiratory pathology from lung auscultation in 572 pediatric outpatients across 5 countries. NPJ Digit Med. 2023;6(1):104. doi: 10.1038/s41746-023-00838-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.LeCun Y., Bengio Y., Hinton G. Deep learning. Nature. 2015;521(7553):436–444. doi: 10.1038/nature14539. [DOI] [PubMed] [Google Scholar]
  • 34.Carney L.H. Speeding up machine hearing. Nature Machine Intelligence. 2021;3(3):190–191. [Google Scholar]
  • 35.Lyon R.F. Cambridge University Press; 2017. Human and Machine Learning: Extracting Meaning from Sound. [Google Scholar]
  • 36.Lillicrap T.P., Santoro A., Marris L., Akerman C.J., Hinton G. Backpropagation and the brain. Nat Rev Neurosci. 2020;21(6):335–346. doi: 10.1038/s41583-020-0277-3. [DOI] [PubMed] [Google Scholar]
  • 37.Kraman S.S. Transmission of lung sounds through light clothing. Respiration. 2008;75(1):85–88. doi: 10.1159/000098404. [DOI] [PubMed] [Google Scholar]
  • 38.Haykin S., Chen Z. The cocktail party problem. Neural computation. 2005;17(9):1875–1902. doi: 10.1162/0899766054322964. [DOI] [PubMed] [Google Scholar]
  • 39.Fu D., Weber C., Yang G., Kerzel M., Nan W., Barros P., et al. What can computational models learn from human selective attention? A review from an audiovisual unimodal and crossmodal perspective. Front Integr Neurosci. 2020;14:10. doi: 10.3389/fnint.2020.00010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Price C.N., Moncrieff D. Defining the role of attention in hierarchical auditory processing. Audiol Res. 2021;11(1):112–128. doi: 10.3390/audiolres11010012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Wodicka G.R., Stevens K.N., Golub H.L., Shannon D.C. Spectral characteristics of sound transmission in the human respiratory system. IEEE Trans Biomed Eng. 1990;37(12):1130–1135. doi: 10.1109/10.64455. [DOI] [PubMed] [Google Scholar]
  • 42.Sfayyih A.H., Sabry A.H., Jameel S.M., Sulaiman N., Raafat S.M., Humaidi A.J., et al. Acoustic-based deep learning architectures for lung disease diagnosis: a comprehensive overview. Diagnostics. 2023;13(10):1748. doi: 10.3390/diagnostics13101748. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Pramono R.X.A., Bowyer S., Rodriguez-Villegas E. Automatic adventitious respiratory sound analysis: a systematic review. PloS One. 2017;12(5) doi: 10.1371/journal.pone.0177926. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Jácome C., Ravn J., Holsbø E., Aviles-Solis J.C., Melbye H., Ailo Bongo L. Convolutional neural network for breathing phase detection in lung sounds. Sensors. 2019;19(8):1798. doi: 10.3390/s19081798. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Skalicky D., Koucky V., Hadraba D., Viteznik M., Dub M., Lopot F. Detection of respiratory phases in a breath sound and their subsequent utilization in a diagnosis. Appl Sci. 2021;11(14):6535. [Google Scholar]
  • 46.Pessoa D., Rocha B.M., Strodthoff C., Gomes M., Rodrigues G., Petmezas G., et al. BRACETS: Bimodal repository of auscultation coupled with electrical impedance thoracic signals. Comput Methods Programs Biomed. 2023;240 doi: 10.1016/j.cmpb.2023.107720. [DOI] [PubMed] [Google Scholar]
  • 47.Garcia-Mendez J.P., Lal A., Herasevich S., Tekin A., Pinevich Y., Lipatov K., et al. Machine learning for automated classification of abnormal lung aounds obtained from public satabases: a systematic review. Bioengineering. 2023;10(10):1155. doi: 10.3390/bioengineering10101155. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Bigelow J., Poremba A. Achilles’ ear? Inferior human short-term and recognition memory in the auditory modality. PloS One. 2014;9(2) doi: 10.1371/journal.pone.0089914. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Goode R.L. Middle ear function, biologic variation, and otosurgical alchemy: Can we turn tin ears into gold? Arch Otolaryngol Head Neck Surg. 1986;112(9):923–924. doi: 10.1001/archotol.1986.03780090019001. [DOI] [PubMed] [Google Scholar]
  • 50.Kiyokawa H., Greenberg M., Shirota K., Pasterkamp H. Auditory detection of simulated crackles in breath sounds. Chest. 2001;119(6):1886–1892. doi: 10.1378/chest.119.6.1886. [DOI] [PubMed] [Google Scholar]
  • 51.Hoeting N.M., McCracken C.E., McConnell M., Sallee D., Iannucci G.J., Oster M.E. Systolic ejection click versus split first heart sound: are our ears deceiving us? Congenit Heart Dise. 2017;12(4):417–420. doi: 10.1111/chd.12460. [DOI] [PubMed] [Google Scholar]
  • 52.Grand View Research Stethoscope market size, share & trends analysis report by technology type (electronic/digital stethoscope, smart stethoscope) by sales channel, by region and segment forecasts, 2024-2030. https://www.grandviewresearch.com/industry-analysis/stethoscopes-market
  • 53.Azmeen A., Vakilzadian H., Haider H., Mathers D.H., Zimmerman R., Bedi S., et al. Heart sounds: Past, present, and future from a technological and clinical perspective–a systematic review. Proc Inst Mech Eng H: J Eng Med. 2023;237(6):669–682. doi: 10.1177/09544119231172858. [DOI] [PubMed] [Google Scholar]
  • 54.Jani V., Danford D.A., Thompson W.R., Schuster A., Manlhiot C., Kutty S. The discerning ear: cardiac auscultation in the era of artificial intelligence and telemedicine. Eur Heart J Digit Health. 2021;2(3):456–466. doi: 10.1093/ehjdh/ztab059. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Arjoune Y., Nguyen T.N., Salvador T., Geggel R.L., May J.W., Pillai D.K., et al. StethAid: a digital auscultation platform for pediatrics. Sensors. 2023;23(12):5750. doi: 10.3390/s23125750. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Roff M., Slifirski O., Grooby E., Marzbanrad F., Malhotra A. Digital stethoscope use in neonates: a systematic review. Newborn. 2023;(3):235–243. [Google Scholar]
  • 57.Grzywalski T., Belluzzo R., Piecuch M., Szajek M., Breborowicz A., Pastusiak A., et al. In: Riaño D., Wilk S., ten Teije A., editors. Vol. 11526. AIME; 2019. Fully interactive lungs auscultation with AI enabled digital stethoscope; pp. 31–35. (Artificial Intelligence in Medicine). (23-26 September 2019) [Google Scholar]
  • 58.Aviles-Solis J.C., Storvoll I., Vanbelle S., Melbye H. The use of spectrograms improves the classification of wheezes and crackles in an educational setting. Sci Rep. 2020;10(1):8461. doi: 10.1038/s41598-020-65354-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Melbye H., Garcia-Marcos L., Brand P., Everard M., Priftis K., Pasterkamp H. Wheezes, crackles and rhonchi: simplifying description of lung sounds increases the agreement on their classification: a study of 12 physicians' classification of lung sounds from video recordings. BMJ Open Respir Res. 2016;3(1) doi: 10.1136/bmjresp-2016-000136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Pasterkamp H. The highs and lows of wheezing: a review of the most popular adventitious lung sound. Pediatr Pulmonol 218;53(2):243-254. [DOI] [PubMed]
  • 61.Dianat B., La Torraca P., Manfredi A., Cassone D., Vacchi C., Sebastiani M., et al. Classification of pulmonary sounds through deep learning for the diagnosis of interstitial lung diseases secondary to connective tissue diseases. Comput Biol Med. 2023;160 doi: 10.1016/j.compbiomed.2023.106928. [DOI] [PubMed] [Google Scholar]
  • 62.Garrote-Corral S., Silva-Fernández L., Seoane-Mato D., Guerra-Rodríguez M., Aburto M., Castaneda S., et al. Screening of interstitial lung disease in patients with rheumatoid arthritis: a systematic review. Reumatol Clín (Engl Ed). 2022;18(10):587–596. doi: 10.1016/j.reumae.2021.07.002. [DOI] [PubMed] [Google Scholar]
  • 63.Pereira C.A., Soares M.R., Boaventura R., Castro M.D.C., Gomes P.S., Gimenez A., et al. Squawks in interstitial lung disease prevalence and causes in a cohort of one thousand patients. Medicine. 2019;98(29) doi: 10.1097/MD.0000000000016419. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64.Forgacs P. Crackles and wheezes. Lancet. 1967;290(7508):203–205. doi: 10.1016/s0140-6736(67)90024-4. [DOI] [PubMed] [Google Scholar]
  • 65.Andersen S., Davidsen A.H., Schirmer H., Melbye H., Spigt M., Aviles-Solis J.C. Interrater and intrarater agreement on heart murmurs. Scand J Prim Health Care. 2022;40(4):491–497. doi: 10.1080/02813432.2022.2159204. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Mosqueira-Rey E., Hernández-Pereira E., Alonso-Ríos D., Bobes-Bascarán J., Fernández-Leal Á. Human-in-the-loop machine learning: a state of the art. Artificial Intelligence Review. 2023;56(4):3005–3054. [Google Scholar]
  • 67.Lucy F.K., Suha K.T., Dipty S.T., Wadud M.S.I., Kadir M.A. Video based non-contact monitoring of respiratory rate and chest indrawing in children with pneumonia. Physiol Meas. 2021;42(10) doi: 10.1088/1361-6579/ac34eb. [DOI] [PubMed] [Google Scholar]
  • 68.Masuda N., Furukawa K., Yairi I.E. Stetho Touch: touch action recognition system by deep learning with stethoscope acoustic sensing. Journal of Information Processing. 2022;30:718–728. [Google Scholar]
  • 69.Woywodt A., Herrmann A., Kielstein J., Haller H., Haubitz M., Purnhagen H. A novel multimedia tool to improve bedside teaching of cardiac auscultation. Postgrad Med J. 2004;80(944):355–357. doi: 10.1136/pgmj.2003.014944. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70.Emeryk A., Derom E., Janeczek K., Kuznar-Kaminska B., Zelent A., Lukaszyk M., et al. Home monitoring of asthma exacerbations in children and adults with use of an AI-aided stethoscope. Ann Fam Med. 2023;21(6):517–525. doi: 10.1370/afm.3039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 71.Elphick H., Ritson S., Rodgers H., Everard M. When a “wheeze” is not a wheeze: acoustic analysis of breath sounds in infants. Eur Respir J. 2000;16(4):593–597. doi: 10.1034/j.1399-3003.2000.16d04.x. [DOI] [PubMed] [Google Scholar]
  • 72.Bohadana A., Azulai H., Jarjoui A., Kalak G., Izbicki G. Influence of observer preferences and auscultatory skill on the choice of terms to describe lung sounds: a survey of staff physicians, residents and medical students. BMJ Open Respir Res. 2020;7(1) doi: 10.1136/bmjresp-2020-000564. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73.Ahmed S., Sultana S., Khan A.M., Islam M.S., Habib G.M., McLane I.M. Digital auscultation as a diagnostic aid to detect childhood pneumonia: a systematic review. J Glob Health. 2022;12 doi: 10.7189/jogh.12.04033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 74.Reinhart R.A. “The report of my death...”: exaggerated predictions on the demise of the stethoscope. Chest. 2022;162(4):872–877. doi: 10.1016/j.chest.2022.05.016. [DOI] [PubMed] [Google Scholar]
  • 75.Zainal H., Xin X., Thumboo J., Fong K.Y. Medical school curriculum in the digital age: perspectives of clinical educators and teachers. BMC Med Educ. 2022;22(1):1–10. doi: 10.1186/s12909-022-03454-z. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from CHEST Pulmonary are provided here courtesy of Elsevier

RESOURCES