Construction and validation of an automatic segmentation method for respiratory sound time labels

Construction and validation of an automatic segmentation method for respiratory sound time labels | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Construction and validation of an automatic segmentation method for respiratory sound time labels Jian Fan, Haoran Ni, Xiulan Chen, Yulin Duan, Wanmin Wang, Fan Xu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5324173/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background :In the field of respiratory system diseases, the utilization of respiratory sounds in auscultation plays a crucial role in the specific disease diagnosis. However, during the process of auscultation, the personal experiences and environmental factors may affect the decision making, leading to diagnostic errors. Therefore, to accurately and effectively obtaining and analyzing respiratory sounds can be positively contribute to the diagnosis and treatment of respiratory system diseases. Objectives :Our aim was to develop an analytical method for the visualization and digitization of respiratory audio data, and to validate its capability to differentiate between various background diseases. Methods :This study collected the respiratory sounds of patients admitted to the Department of General Medicine of Shanghai Changhai Hospital from June to December 2023. After strict screening according to the inclusion and exclusion criteria, a total of 84 patients were included. The research process includes using an electronic stethoscope to collect lung sounds from patients in a quiet environment. The patients expose their chests and lie flat. Sound data is collected at six landmark positions on the chest. The collected audio files are imported into an analysis tool for segmentation and feature extraction. Specific analysis methods include distinguishing heart sounds and respiratory sounds, segmenting respiratory sounds, determining the inspiratory and expiratory phases, and using a tool developed by the team for automatic segmentation encoding. Results :We standardized the respiratory sounds of 84 patients and segmented multiple respiratory cycles. Following the localization and segmentation of the respiratory cycles based on label information, we calculated the average and standard deviations of the amplitude features for each segment of the respiratory cycle. The results indicated differences among various diseases. Conclusions :The robust algorithm platform is capable to segmenting the respiratory sounds into inhale and exhale phase accordingly, then to comparing the difference between different background disease. This method provides objective evidence for auscultation of respiratory sounds and visual display of breath sounds. Respiratory sound Artificial intelligence Digital auscultation Automatic segmentation method Respiratory system diseases Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction The physiological respiratory cycle can be divided into the inhalation and exhalation phase. Inhalation causes the expansion of the lungs, allowing air to be drawn in. While, the exhalation causes the contraction of the lungs, leading to the expulsion of air. The respiratory sound signal can reflect the valuable information about the physiological or pathological status of the respiratory system [ 1 ]. For instance, the abnormalities of expiratory phase are frequently observed in conditions such as asthma, chronic obstructive pulmonary disease (COPD), lung infections, bronchiectasis, and interstitial lung disease. Moreover, the abnormalities in inspiratory phase commonly manifest in upper respiratory tract obstruction, muscle weakness, pleural effusion, and pneumothorax. Currently, lung auscultation is widely recognized as the most robust and reliable approach for detecting pulmonary abnormalities. For example, Xu et al. proposed an innovative approach to accurately assess lung function by analyzing cough sounds recorded via mobile devices. This method overcomes limitations of time and location, enabling frequent sampling and evaluation of lung function in the home environment using smartphones [ 2 ] Furthermore, Raj V et al. investigated the profound potential of utilizing lung auscultation, specifically vesicular and bronchial breath sounds, for diagnosing and addressing COVID-19 concerns. These findings hold significant implications in the realm of medical research [ 3 ]. However, the auscultation of the lungs highly depends on the proficient respiratory specialists and the technique relies on the acoustic manifestations of the respiratory cycle (i.e., contraction and relaxation of the respiratory muscles) to detect abnormalities in lung function. These important respiratory sounds, encompassing both the inspiratory and expiratory phases, are localized at six distinct auscultation sites on the chest surface. However, they are highly susceptible to interference throughout the respiratory cycle, potentially leading to missed detection of vital respiratory signals. Consequently, there is an urgent imperative to explore enhanced methodologies for analyzing respiratory sounds. Thanks to the rapid development of artificial intelligence, it enables us to capture more fleeting signals in lung disease [ 4 – 6 ]. Several algorithms have been validated the robust automation capabilities in the segmentation of respiratory sounds. For example, McLane et al. proposed a technique for the detection and classification of respiratory sounds using vesicular sounds and crackle peak detection [ 7 ]. Through quantitative analysis of respiratory sounds, clinicians can differentiate between different types of moist rales, determine their timing within the breathing cycle, and assess their severity. The study conducted by Huang et al. employed an artificial intelligence in terms of deep learning methodology to transform lung sounds into two-dimensional spectrograms and used convolutional neural networks for end-to-end recognition of respiratory system diseases or abnormal lung sounds [ 8 ].To achieve segmentation results that align with the sequential nature of respiratory sounds, the author proposes a Python Package called TSSEARCH, which integrates subsequence search and time series similarity measurement with sequence timing models [ 9 ]. The collected respiratory sounds can be objectively analyzed through computerized intelligent analysis, allowing the signals to be decomposed into different intensities and frequencies for a more accurate characterization of such signals. This technology enables a more objective analysis of respiratory sounds and noise to better understand them [ 10 ].Techniques used in other applications include the determination of breath sound boundaries, feature extraction of breath sounds and noise, differentiation between expiratory and inspiratory phases, and reduction of noise interference. To accurately characterize the parameters of the breath sound signal, artificial intelligence algorithms make the whole decipher process more intelligent. Complex network analysis can be applied to the study of bioacoustic signals and bronchial breath sounds. By utilizing machine learning techniques that extract graph features, the intensity and frequency of respiratory sounds can be measured, which represent fundamental characteristics of sound during the breathing process [ 11 ].In addition, the electronic stethoscope was used in a quiet environment with full patient cooperation to digitally record respiratory sounds, it providing a solid foundation for the analysis of raw data [ 12 ]. The present study introduced a novel approach for the automated segmentation of respiratory sounds and extraction of fundamental frequency and intensity parameters. This methodology combines the Audio Data Analysis Tool with customized MATLAB code. The proposed method represents a significant advancement in the field of artificial intelligence analysis of respiratory sounds, making valuable contributions to the diagnosis and prediction of respiratory system diseases. Methods General information Collect respiratory sounds of 84 patients admitted to the Department of General Medicine at Shanghai Changhai Hospital from June 2023 to December 2023. Inclusion and exclusion criteria Inclusion Criteria: 18 years of age or older; agree the consent to the collection of respiratory sounds. Exclusion criteria included: unwillingness to cooperate with respiratory sound collection during the research period; presence of congenital airway abnormalities, pulmonary underdevelopment, or other related illnesses; severe substantial organ damage preventing cooperation; currently receiving respiratory machine treatment; presence of contagious respiratory diseases. Research Process Firstly, ETZ-1A(C) electronic stethoscope was used to collect lung sounds from patients while maintaining a quiet environment. The patient was directed to assume a supine position on the bed, with their chest exposed. Before use, disinfect and warm the stethoscope. Place the auscultation head on the desired listening area, apply appropriate pressure, and start recording on the mobile phone app. The duration of each recording should cover at least 2 respiratory cycles in each lung auscultation zone. The auscultation of lung sounds is performed at six different locations. These locations are as follows: right upper chest (intersection between the second rib and the midclavicular line on the right side), left upper chest (intersection between the second rib and the midclavicular line on the left side), right middle chest (intersection between the fourth rib and the anterior axillary line on the right side), left middle chest (intersection between the fourth rib and the anterior axillary line on the left side), right lower chest (intersection between the sixth rib and the anterior axillary line on the right side), and left lower chest (intersection between the sixth rib and the anterior axillary line on the left side). It is important to avoid auscultating over areas where heart sounds are heard. Data from each participant was collected continuously for three days, with two sessions per day (8:00–11:00 and 14:30 − 17:00). The best audio data from one of the three days was selected for analysis. Finally, the audio files imported into audio data analysis tool and automatically segment the audio data representing one cycle into two segments (inspiratory phase S1 and expiratory phase S2).Then, the detail feature parameters were extracted from each frame of the lung sound signal, including time-domain features (such as energy, mean value, peak value), frequency-domain features (such as spectral energy, frequency peak), and time-frequency domain features (such as wavelet coefficients). The differentiation of cardiac and respiratory sounds At the beginning, to denoise the background noise from heartbeat, including uniform the sampling rate and normalization processing. Then the Short-Time Fourier Transform (STFT) was used to convert it into a frequency spectrum representation. The STFT amplitude spectrum was processed by a Recursive Neural Network (RNN) model to obtain time-frequency masks for the heart sound signal and noise signal. The STFT magnitude spectrum was multiplied by the mask output derived from the RNN, followed by signal reconstruction using inverse STFT to obtain lung sounds contaminated with noise. Subsequently, the heart sound signal was filtered and subjected to Hilbert transform for feature extraction. The heart rate and duration were calculated, and a Hidden Markov Model (HMM) was employed to obtain different sub-state sequences of the heart sound signal. After denoising and concatenating the paired state sequences, iterative iterations were performed until the set number of iterations was reached, ultimately achieving effective separation and reconstruction of noisy heart and lung sound signals. The segmentation of respiratory sounds The Audio Data Analysis Tool software was utilized to visualize the respiratory sounds, ensuring that the starting point in the audio file aligns with either inhalation or exhalation phase of breath sounds. This approach effectively eliminates the interference from breathing and external noise. The specific characteristics of sound were taken into consideration, and feature parameters were extracted from each frame of lung sound signals. These parameters include temporal features (duration), spectral features (spectral energy, frequency peak, spectral slope), and intensity features (maximum amplitude, minimum amplitude, skewness, kurtosis). These results were examined to identify the most consistent and distinguishable respiratory sound signals for segmentation. The segments of the inspiratory and expiratory phases A breathing cycle consists of an inhalation phase and an exhalation phase, where the inhalation phase is defined as the time from the start of S1 to the beginning of S2. The exhalation phase is defined as the time from the beginning of S2 to the start of the next breathing cycle at S1. The relevant audio clips were being trimmed, and the audio file was segmented into distinct sections. By executing customized MATLAB code, the detail parameters pertaining to inhalation and exhalation phases can be extracted. The segmentation of inspiratory and expiratory phases The amplitude and time axis of the S1 peak were automatically determined by our audio analysis data tool, based on the baseline at the end of preprocessing. Additionally, it calculated the peak value of S1 and identifies the frequency range preceding the valley of the previous wave. The start and end times of S1 can be determined in this manner. Subsequently, the second peak occurring between two adjacent S1 peaks was utilized as a distinctive feature for extracting the S2 peak. The S2 peaks that significantly deviated from the average value were filtered out, followed by selecting the data with the highest statistical significance. The corresponding time of this selected data serves as a label to obtain the starting point and ending point of S2. The starting times of S1 and S2 were subsequently calculated. Finally, utilizing statistical data, the start and end times, states, decibels, amplitudes, and other relevant information for each state were computed. Automatic segmentation encoding The Respiratory Data Analysis Tool, developed by our team, employs MATLAB for comprehensive audio analysis and feature extraction. It also facilitates automated generation of spectrograms and provides visual representations, preprocessed audio samples, as well as detailed data results. Results Population dataset A total of 84 patients were recruited, including 43male individuals with an average age of (65.50&thifnsp;± 9.45) years and 41 female individuals with an average age of (64.90 ± 6.48) years. The details of demographics of the subjects refer the Table 1 . Table 1 Demographics of subjects. Characteristic Male (mean ± std) Female (mean ± std) Number 43 41 Heart rate 71.60 ± 4.20 68.80 ± 6.40 Blood oxygen 96.38 ± 1.23 98.00 ± 0.82 Age 65.50 ± 9.45 64.90 ± 6.48 Height 176.12 ± 5.32 159.78 ± 4.10 Weight 69.32 ± 9.86 50.78 ± 6.12 BMI 22.38 ± 3.26 19.92 ± 2.50 Visualization and denoise the respiratory sound Before standardization, the collected respiratory sounds were difficult to identify due to interference from heart sounds in the three auscultation areas on the left side, which were close to the heart (Fig. 2 ). In contrast, the three auscultation areas on the right side experienced less interference from heart sounds (Fig. 3 ). After undergoing standardization processing, the characteristics of respiratory sounds became evident, resulting in noise reduction and signal amplification effects, leading to clearer waveforms (Fig. 4 ). Respiratory sound cycle segmentation The segmentation of multiple respiratory cycles from an audio file was based on the localization of two adjacent "similar images" in phonocardiography (PCG)(defined as two consecutive images with similar amplitude and duration) (Fig. 5 ). The inhalation phase (S1) and exhalation phase (S2) separation After positioning and segmenting the respiratory cycle, audio data analysis tools can identified the intensity, duration, and frequency domain characteristics of S1 and S2 as follows: (I) S1 has a longer duration than S2 and its amplitude was significantly higher than that of S2. (II) Both S1 and S2 have a certain frequency range in physiological terms. (III) In PCG, S2 occurs between adjacent S1 signals. By combining these features, we determined the start and end positions of both S1 and S2 (Fig. 5 ). Segmentation of inhalation and exhalation phases based on label information. The time tag information was extracted from the Respirtory Data Analysis Tool to accurately separate the inhalation and exhalation periods. After identifying and locating S1 and S2, label values were calculated (Fig. 6 ) to precisely clip the recordings. The recorded respiratory sounds include precise start and end times for each breathing cycle. The amplitude of respiratory sounds can indicate the level of physical activity in the respiratory muscles, and previous studies suggested that this could be valuable for COVID-19 detection [ 13 ]. Sound magnitude was utilized to calculate and filter loudness values, specifically determining the amplitudes of different segments (S1, S2) within the respiratory cycle. In this study, we computed the average and standard deviation of amplitude features (Maxdb, Mindb, Meandb, Midledb) for each segment of the respiratory cycle. Spectrum Spectrum analysis is the most widely used method in respiratory sound analysis. In this study, MATLAB code (Supplementary data) was used to perform formula calculations on two segments (S1 and S2). Cross-validation based on the respiratory cycle. In order to validate the accuracy of our segmentation results, the double-blind validation also conducted respiratory phase verification with experienced experts in the field of respiratory diseases. This was performed to record the subjects' respiratory rates during the data collection process. Discussion Respiratory system diseases are one of the most common health issues affecting people in their daily lives. With a high incidence rate and limited diagnostic methods, there is an urgent need for simple, reliable, and radiation-free diagnostic techniques. Intelligent auscultation technology for respiratory sounds can play a significant role in predicting respiratory system diseases [ 14 – 15 ]. However, during the performing respiratory auscultation in a clinical frontline, even experienced respiratory specialists can be disturbed in the diagnosis of respiratory disorders due to the potential influence of a variety of external factors, which may lead to misdiagnosis or omission. The artificial intelligence recognition technology is capable of detecting acoustic information from respiratory sound signals, encompassing temporal domain features (e.g., energy, average value, peak value), frequency domain features (e.g., spectral energy, frequency peak), and time-frequency domain features (e.g., wavelet coefficients). Moreover, this method presented the advantages of batch processing and convenience in screening and diagnosing respiratory system diseases [ 16 ].In this study, we developed and validated an automated method for segmenting respiratory sounds and extracted basic acoustic information and sound intensity parameters using our audio data analysis tool. Through data analysis, no significant differences were found in respiratory sound against the respiratory system diseases between detected by the audio data analysis tool and radiological examinations. Diagnosis of respiratory diseases can be greatly improved by utilizing intelligent digital auscultation technology. McLane et al. presented a comprehensive signal analysis method for analyzing challenging recordings of crackles, which includes five processing modules: (1) motion artifact detection, (2) deep learning denoising network, (3) respiratory cycle segmentation, (4) separation of discontinuous non-tonal components from bubble sounds, and (5) crackle peak detection. These methods provide quantifiable analysis of respiratory sounds, enabling clinicians to differentiate various types of crackles based on their timing and severity within the respiratory cycle [ 17 ].To address the occasional misidentification of respiratory sounds by medical professionals, Kim et al. developed an automatic classification system for respiratory sounds by using of deep learning convolutional neural networks (CNN). They successfully categorized 1918 types of respiratory sounds from clinical records into various categories including normal, crackles, wheezes, and rhonchi. By integrating a pre-trained sequential image feature extractor with the respiratory sound data and CNN classifier, they established a predictive model for accurately classifying respiratory sounds. This innovative model holds great potential in facilitating prompt diagnosis and appropriate treatment of respiratory system diseases [ 18 ].Heitmann et al. used 35.9 hours of auscultation audio data from 572 pediatric outpatient patients to develop Deep Breath: a deep learning model for identifying sound features of acute respiratory diseases in children. The model consists of a convolutional neural network and a logistic regression classifier, which aggregated estimates recorded from eight chest locations into a single prediction at the patient level. Internal 5-fold cross-validation was performed on the results, and external validation was conducted in three other countries (Senegal, Cameroon, Morocco). Deep Breath provides an interpretable framework for deep learning to identify objective audio features of respiratory diseases [ 19 ].The major of existing lung sound recognition methods neglect the correlation between temporal and spectral information of lung sounds, resulting in insufficient capturing of detailed features. Shi et al. introduced a model that incorporates wavelet feature enhancement and time-frequency synchronized modeling techniques. This model comprises a dual wavelet analysis module (DWAM), a cubic network, and an attention module. Experimental results on both merged datasets as well as those from the 2017 International Conference on Biomedical and Health Informatics demonstrate that their proposed framework outperforms existing models by over 1.36% and 4.28%, respectively [ 20 ]. Our research shares similarities with the theirs, commencing from the rigorous selection of participants based on specific inclusion and exclusion criteria, employing advanced electronic stethoscopes for capturing respiratory sound data, followed by preprocessing of sound signals, extraction of relevant features, and ultimately achieving comprehensive analysis of respiratory sounds. However, it has been suggested in certain studies that harnessing machine learning models for analyzing lung sounds can further explore their diagnostic potential for other diseases related to the respiratory system [ 21 ]. The innovation in our respiratory sound signal processing embedded in the utilization of our self-developed software, Repertory Data Analysis Tool, for data processing. This tool effectively aggregates time labels for respiratory sounds and employs MATLAB for audio analysis, feature extraction, and the generation of partial spectrograms. In addition, all acoustic parameters can be collected, which also facilitates researchers to diversify the data. Conclusion The integration of electronic stethoscope and artificial intelligence technology has facilitated the digital auscultation technique, enabling the digitized collection of respiratory sounds and precise identification of inspiratory phase (S1) and expiratory phase (S2). The analysis results obtained through artificial intelligence technology on respiratory sounds demonstrate robust scientific stability. This approach provides objective evidence for respiratory auscultation and noise assessment, thus holding significant importance in clinical research. Limitations Despite our efforts to minimize noise during data collection, significant amounts of noise were present due to uncontrollable factors such as heart sounds, poor stethoscope fit on thin patients, and external environmental factors. Additionally, manually cutting audio files at their beginnings and endings during sound processing could potentially affect data analysis results. Declarations Ethics approval and consent to participate The ethics committee at Ethics Committee of Shanghai Changhai Hospital approved the investigation (CHEC2024-049). The written informed consent has been obtained from all participants. Consent for publication Not applicable. Availability of data and materials The author will provide the original data supporting the conclusions of this article without any unnecessary withholding. Competing interests The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Funding The author(s) declare that financial support was received for the research, authorship, and/or publication of this article. This work was supported by Special Fund for “ Research on Community Medicine and Health Management” in Shanghai (2023SQ01). Shanghai "Rising Stars of Medical Talent" Youth Development Program （ Youth Medical Talents – General Practitioner Program ） Authors' contributions FJ, NHR, CXL: Data collection management, Formal analysis, Resources, Ethical application, Writing original draft. DYL: Formal analysis, Validation, Project management, Resources, Writing original draft. WWM : Data collection, Data classification management. XF, SY: Methodology, Grant acquisition, Project management, Review and editing. Acknowledgements Not applicable. References McLane I, Lauwers E, Stas T, et al. Comprehensive Analysis System for Automated Respiratory Cycle Segmentation and Crackle Peak Detection. IEEE J Biomed Health Inform 2022; 26(4): 1847–1860 Xu W, He G, Pan C, et al. A forced cough sound based pulmonary function assessment method by using machine learning. Front Public Health 2022; 10: 1015876 Raj V, Renjini A, Swapna MS, Sreejyothi S, Sankararaman S. Nonlinear time series and principal component analyses: Potential diagnostic tools for COVID-19 auscultation. Chaos Solitons Fractals 2020; 140: 110246 Chao HS, Tsai CY, Chou CW, et al. Artificial Intelligence Assisted Computational Tomographic Detection of Lung Nodules for Prognostic Cancer Examination: A Large-Scale Clinical Trial. Biomedicines 2023; 11(1): 147 Qi J, Hong B, Tao R, et al. Prediction model for malignant pulmonary nodules based on cfMeDIP-seq and machine learning. Cancer Sci 2021; 112(9): 3918–3923 Chen M, Copley SJ, Viola P, Lu H, Aboagye EO. Radiomics and artificial intelligence for precision medicine in lung cancer treatment. Semin Cancer Biol 2023; 93: 97–113 Huang DM, Huang J, Qiao K, Zhong NS, Lu HZ, Wang WJ. Deep learning-based lung sound analysis for intelligent stethoscope. Mil Med Res 2023; 10(1): 44 D Folgado MB, M Antunes MLN, Liu H. Tssearch: Time series subsequence search library. SoftwareX 2022; ZF Bertrand KDS, Sánchez DI. Lung auscultation in the 21th century. Revista Chilena de … V Raj MSS, Sankararaman S. Bioacoustic signal analysis through complex network features. Comput Biol Med 2022; Chen Z, Li M, Wang R, et al. Diagnosis of COVID-19 via acoustic analysis and artificial intelligence by monitoring breath sounds on smartphones. J Biomed Inform 2022; 130: 104078 Lee SH, Kim YS, Yeo MK, et al. Fully portable continuous real-time auscultation with a soft wearable stethoscope designed for automated disease diagnosis. Sci Adv 2022; 8(21): eabo5867 Rahman T, Ibtehaz N, Khandakar A, et al. QUCoughScope: An Intelligent Application to Detect COVID-19 Patients Using Cough and Breath Sounds. Diagnostics (Basel). 2022;12(4):920. Published 2022 Apr 7. doi: 10.3390/diagnostics12040920 Gupta P, Moghimi MJ, Jeong Y, Gupta D, Inan OT, Ayazi F. Precision wearable accelerometer contact microphones for longitudinal monitoring of mechano-acoustic cardiopulmonary signals. NPJ Digit Med 2020; 3: 19 Kim Y, Hyon Y, Jung SS, et al. Respiratory sound classification for crackles, wheezes, and rhonchi in the clinical field using deep learning. Sci Rep 2021; 11(1): 17186 Heitmann J, Glangetas A, Doenz J, 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 Shi L, Zhang Y, Zhang J. Lung Sound Recognition Method Based on Wavelet Feature Enhancement and Time-Frequency Synchronous Modeling. IEEE J Biomed Health Inform 2023; 27(1): 308–318 Dar JA, Srivastava KK, Mishra A. Lung anomaly detection from respiratory sound database (sound signals). Comput Biol Med 2023; 164: 107311 Heitmann J, Glangetas A, Doenz J, 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. Published 2023 Jun 2. doi: 10.1038/s41746-023-00838-3 Shi L, Zhang Y, Zhang J. Lung Sound Recognition Method Based on Wavelet Feature Enhancement and Time-Frequency Synchronous Modeling. IEEE J Biomed Health Inform. 2023;27(1):308–318. doi: 10.1109/JBHI.2022.3210996 Dar JA, Srivastava KK, Mishra A. Lung anomaly detection from respiratory sound database (sound signals). Comput Biol Med. 2023;164:107311. doi: 10.1016/j.compbiomed.2023.107311 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5324173","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":372779551,"identity":"c05fd44d-c358-42de-befd-3d2e27c37b8f","order_by":0,"name":"Jian Fan","email":"","orcid":"","institution":"Department of General Practice, The First Hospital Affiliated of Naval Medical University, Changhai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jian","middleName":"","lastName":"Fan","suffix":""},{"id":372779553,"identity":"9d57ad79-755d-4899-bc16-841edbf249b6","order_by":1,"name":"Haoran Ni","email":"","orcid":"","institution":"Department of Respiratory and Critical Care, The First Hospital Affiliated of Naval Medical University, Changhai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Haoran","middleName":"","lastName":"Ni","suffix":""},{"id":372779554,"identity":"0ac34d79-02d8-4744-af9f-b9cc45cdb24e","order_by":2,"name":"Xiulan Chen","email":"","orcid":"","institution":"Daning Road Community Health Service Center","correspondingAuthor":false,"prefix":"","firstName":"Xiulan","middleName":"","lastName":"Chen","suffix":""},{"id":372779555,"identity":"93d1e83f-8e53-42ca-805c-701190bb0727","order_by":3,"name":"Yulin Duan","email":"","orcid":"","institution":"Department of Public Health, Chengdu Medical College","correspondingAuthor":false,"prefix":"","firstName":"Yulin","middleName":"","lastName":"Duan","suffix":""},{"id":372779557,"identity":"fd2dd2d2-ffa3-414e-b262-821489fd0bfe","order_by":4,"name":"Wanmin Wang","email":"","orcid":"","institution":"Department of General Practice, The First Hospital Affiliated of Naval Medical University, Changhai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wanmin","middleName":"","lastName":"Wang","suffix":""},{"id":372779561,"identity":"bbac9748-1f19-4ff8-905e-5c5a735ba285","order_by":5,"name":"Fan Xu","email":"","orcid":"","institution":"Department of Public Health, Chengdu Medical College","correspondingAuthor":false,"prefix":"","firstName":"Fan","middleName":"","lastName":"Xu","suffix":""},{"id":372779563,"identity":"7ecc4437-2cbb-4250-abd5-da126b15d374","order_by":6,"name":"Yan Shang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBElEQVRIiWNgGAWjYDACCShtf/9g44MEAxsefv4GIrUw3GBuNvhQkSYjOeMA0VrY2yRnnDlsY9CQgF+H/OzmZw+/ttnlMc5ubJDmbTvPY8BwgPHDxxzcWgzuHDM3lm1LLmaWOdhgzNt2m8ecuYFZcuY2PFokEsykJduYE9sYEhuSQVosGw6wMfPi0SI/I/0bUEt9Yg9Qy2HetnM8BgcS8GthuJFjJvmx7XDiDInExsYZZw4Q1mJwI6dMmuHc8cQNPAebGT5UJPNIzjjYjNcvQIdtk/xRVp24gb39+Y8EAzt7fv7mgx8+4nMYEDDzsqHwGRvwqwcp+fGHoJpRMApGwSgYyQAAWMVY3bzE++oAAAAASUVORK5CYII=","orcid":"","institution":"Department of General Practice, The First Hospital Affiliated of Naval Medical University, Changhai Hospital","correspondingAuthor":true,"prefix":"","firstName":"Yan","middleName":"","lastName":"Shang","suffix":""}],"badges":[],"createdAt":"2024-10-24 08:38:46","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5324173/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5324173/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":68697463,"identity":"52283cba-1196-48b5-87d3-50cbfe4b1201","added_by":"auto","created_at":"2024-11-11 06:57:44","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":153171,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGeneral technical workflow\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/ba44c5d1eea067a6c2d75836.jpeg"},{"id":68698000,"identity":"6b0b7223-9c46-4cc5-8dd6-b0dfe923cc1a","added_by":"auto","created_at":"2024-11-11 07:05:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":48946,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003elung sounds before denoise process.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/b5960e3e94a03bbf332b9462.png"},{"id":68697465,"identity":"f0c5f5e2-1b56-4feb-a730-2263c1c65b6b","added_by":"auto","created_at":"2024-11-11 06:57:44","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":30317,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eLung sounds after denoise process.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/92f45932eb192c6be4143bd6.png"},{"id":68697999,"identity":"b7669915-fdcf-42db-8b36-d71ac30753d2","added_by":"auto","created_at":"2024-11-11 07:05:44","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":24045,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePreprocessed audio recordings of lung sounds.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/70c5821aeceec320e7510cd6.png"},{"id":68697467,"identity":"c046d0ef-f586-4412-9002-dccf7cf259a0","added_by":"auto","created_at":"2024-11-11 06:57:44","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":34175,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBreath cycle segment\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/c6f13efe70b21039dea95324.png"},{"id":68697461,"identity":"4267f814-49d1-4257-8838-955873425618","added_by":"auto","created_at":"2024-11-11 06:57:44","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":34237,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 5. Inhale and exhale separate in every breath cycle\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/8a21ef50156de3bab70d7361.png"},{"id":68697462,"identity":"e1e6f415-4e7d-4fa1-839c-47a278677d70","added_by":"auto","created_at":"2024-11-11 06:57:44","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":135426,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 6. Start and end times of S1 and S2\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/1356371b545af7572dbb83c1.png"},{"id":72956099,"identity":"8e044a02-3f7f-4dab-8e75-07b99fee6120","added_by":"auto","created_at":"2025-01-04 11:01:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1046720,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5324173/v1/dde2eb20-3e71-4c8a-ada9-c4ecb80bc5d2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Construction and validation of an automatic segmentation method for respiratory sound time labels","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe physiological respiratory cycle can be divided into the inhalation and exhalation phase. Inhalation causes the expansion of the lungs, allowing air to be drawn in. While, the exhalation causes the contraction of the lungs, leading to the expulsion of air. The respiratory sound signal can reflect the valuable information about the physiological or pathological status of the respiratory system [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. For instance, the abnormalities of expiratory phase are frequently observed in conditions such as asthma, chronic obstructive pulmonary disease (COPD), lung infections, bronchiectasis, and interstitial lung disease. Moreover, the abnormalities in inspiratory phase commonly manifest in upper respiratory tract obstruction, muscle weakness, pleural effusion, and pneumothorax. Currently, lung auscultation is widely recognized as the most robust and reliable approach for detecting pulmonary abnormalities. For example, Xu et al. proposed an innovative approach to accurately assess lung function by analyzing cough sounds recorded via mobile devices. This method overcomes limitations of time and location, enabling frequent sampling and evaluation of lung function in the home environment using smartphones [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] Furthermore, Raj V et al. investigated the profound potential of utilizing lung auscultation, specifically vesicular and bronchial breath sounds, for diagnosing and addressing COVID-19 concerns. These findings hold significant implications in the realm of medical research [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHowever, the auscultation of the lungs highly depends on the proficient respiratory specialists and the technique relies on the acoustic manifestations of the respiratory cycle (i.e., contraction and relaxation of the respiratory muscles) to detect abnormalities in lung function. These important respiratory sounds, encompassing both the inspiratory and expiratory phases, are localized at six distinct auscultation sites on the chest surface. However, they are highly susceptible to interference throughout the respiratory cycle, potentially leading to missed detection of vital respiratory signals. Consequently, there is an urgent imperative to explore enhanced methodologies for analyzing respiratory sounds.\u003c/p\u003e \u003cp\u003eThanks to the rapid development of artificial intelligence, it enables us to capture more fleeting signals in lung disease [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Several algorithms have been validated the robust automation capabilities in the segmentation of respiratory sounds. For example, McLane et al. proposed a technique for the detection and classification of respiratory sounds using vesicular sounds and crackle peak detection [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Through quantitative analysis of respiratory sounds, clinicians can differentiate between different types of moist rales, determine their timing within the breathing cycle, and assess their severity. The study conducted by Huang et al. employed an artificial intelligence in terms of deep learning methodology to transform lung sounds into two-dimensional spectrograms and used convolutional neural networks for end-to-end recognition of respiratory system diseases or abnormal lung sounds [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].To achieve segmentation results that align with the sequential nature of respiratory sounds, the author proposes a Python Package called TSSEARCH, which integrates subsequence search and time series similarity measurement with sequence timing models [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe collected respiratory sounds can be objectively analyzed through computerized intelligent analysis, allowing the signals to be decomposed into different intensities and frequencies for a more accurate characterization of such signals. This technology enables a more objective analysis of respiratory sounds and noise to better understand them [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].Techniques used in other applications include the determination of breath sound boundaries, feature extraction of breath sounds and noise, differentiation between expiratory and inspiratory phases, and reduction of noise interference. To accurately characterize the parameters of the breath sound signal, artificial intelligence algorithms make the whole decipher process more intelligent.\u003c/p\u003e \u003cp\u003eComplex network analysis can be applied to the study of bioacoustic signals and bronchial breath sounds. By utilizing machine learning techniques that extract graph features, the intensity and frequency of respiratory sounds can be measured, which represent fundamental characteristics of sound during the breathing process [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].In addition, the electronic stethoscope was used in a quiet environment with full patient cooperation to digitally record respiratory sounds, it providing a solid foundation for the analysis of raw data [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe present study introduced a novel approach for the automated segmentation of respiratory sounds and extraction of fundamental frequency and intensity parameters. This methodology combines the Audio Data Analysis Tool with customized MATLAB code. The proposed method represents a significant advancement in the field of artificial intelligence analysis of respiratory sounds, making valuable contributions to the diagnosis and prediction of respiratory system diseases.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eGeneral information\u003c/h2\u003e \u003cp\u003eCollect respiratory sounds of 84 patients admitted to the Department of General Medicine at Shanghai Changhai Hospital from June 2023 to December 2023.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eInclusion and exclusion criteria\u003c/h3\u003e\n\u003cp\u003eInclusion Criteria: 18 years of age or older; agree the consent to the collection of respiratory sounds. Exclusion criteria included: unwillingness to cooperate with respiratory sound collection during the research period; presence of congenital airway abnormalities, pulmonary underdevelopment, or other related illnesses; severe substantial organ damage preventing cooperation; currently receiving respiratory machine treatment; presence of contagious respiratory diseases.\u003c/p\u003e\n\u003ch3\u003eResearch Process\u003c/h3\u003e\n\u003cp\u003eFirstly, ETZ-1A(C) electronic stethoscope was used to collect lung sounds from patients while maintaining a quiet environment. The patient was directed to assume a supine position on the bed, with their chest exposed. Before use, disinfect and warm the stethoscope. Place the auscultation head on the desired listening area, apply appropriate pressure, and start recording on the mobile phone app. The duration of each recording should cover at least 2 respiratory cycles in each lung auscultation zone. The auscultation of lung sounds is performed at six different locations. These locations are as follows: right upper chest (intersection between the second rib and the midclavicular line on the right side), left upper chest (intersection between the second rib and the midclavicular line on the left side), right middle chest (intersection between the fourth rib and the anterior axillary line on the right side), left middle chest (intersection between the fourth rib and the anterior axillary line on the left side), right lower chest (intersection between the sixth rib and the anterior axillary line on the right side), and left lower chest (intersection between the sixth rib and the anterior axillary line on the left side). It is important to avoid auscultating over areas where heart sounds are heard. Data from each participant was collected continuously for three days, with two sessions per day (8:00\u0026ndash;11:00 and 14:30\u0026thinsp;\u0026minus;\u0026thinsp;17:00). The best audio data from one of the three days was selected for analysis.\u003c/p\u003e \u003cp\u003eFinally, the audio files imported into audio data analysis tool and automatically segment the audio data representing one cycle into two segments (inspiratory phase S1 and expiratory phase S2).Then, the detail feature parameters were extracted from each frame of the lung sound signal, including time-domain features (such as energy, mean value, peak value), frequency-domain features (such as spectral energy, frequency peak), and time-frequency domain features (such as wavelet coefficients).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eThe differentiation of cardiac and respiratory sounds\u003c/h3\u003e\n\u003cp\u003eAt the beginning, to denoise the background noise from heartbeat, including uniform the sampling rate and normalization processing. Then the Short-Time Fourier Transform (STFT) was used to convert it into a frequency spectrum representation. The STFT amplitude spectrum was processed by a Recursive Neural Network (RNN) model to obtain time-frequency masks for the heart sound signal and noise signal. The STFT magnitude spectrum was multiplied by the mask output derived from the RNN, followed by signal reconstruction using inverse STFT to obtain lung sounds contaminated with noise. Subsequently, the heart sound signal was filtered and subjected to Hilbert transform for feature extraction. The heart rate and duration were calculated, and a Hidden Markov Model (HMM) was employed to obtain different sub-state sequences of the heart sound signal. After denoising and concatenating the paired state sequences, iterative iterations were performed until the set number of iterations was reached, ultimately achieving effective separation and reconstruction of noisy heart and lung sound signals.\u003c/p\u003e\n\u003ch3\u003eThe segmentation of respiratory sounds\u003c/h3\u003e\n\u003cp\u003eThe Audio Data Analysis Tool software was utilized to visualize the respiratory sounds, ensuring that the starting point in the audio file aligns with either inhalation or exhalation phase of breath sounds. This approach effectively eliminates the interference from breathing and external noise. The specific characteristics of sound were taken into consideration, and feature parameters were extracted from each frame of lung sound signals. These parameters include temporal features (duration), spectral features (spectral energy, frequency peak, spectral slope), and intensity features (maximum amplitude, minimum amplitude, skewness, kurtosis). These results were examined to identify the most consistent and distinguishable respiratory sound signals for segmentation.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eThe segments of the inspiratory and expiratory phases\u003c/h2\u003e \u003cp\u003eA breathing cycle consists of an inhalation phase and an exhalation phase, where the inhalation phase is defined as the time from the start of S1 to the beginning of S2. The exhalation phase is defined as the time from the beginning of S2 to the start of the next breathing cycle at S1. The relevant audio clips were being trimmed, and the audio file was segmented into distinct sections. By executing customized MATLAB code, the detail parameters pertaining to inhalation and exhalation phases can be extracted.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eThe segmentation of inspiratory and expiratory phases\u003c/h3\u003e\n\u003cp\u003eThe amplitude and time axis of the S1 peak were automatically determined by our audio analysis data tool, based on the baseline at the end of preprocessing. Additionally, it calculated the peak value of S1 and identifies the frequency range preceding the valley of the previous wave. The start and end times of S1 can be determined in this manner. Subsequently, the second peak occurring between two adjacent S1 peaks was utilized as a distinctive feature for extracting the S2 peak. The S2 peaks that significantly deviated from the average value were filtered out, followed by selecting the data with the highest statistical significance. The corresponding time of this selected data serves as a label to obtain the starting point and ending point of S2. The starting times of S1 and S2 were subsequently calculated. Finally, utilizing statistical data, the start and end times, states, decibels, amplitudes, and other relevant information for each state were computed.\u003c/p\u003e\n\u003ch3\u003eAutomatic segmentation encoding\u003c/h3\u003e\n\u003cp\u003eThe Respiratory Data Analysis Tool, developed by our team, employs MATLAB for comprehensive audio analysis and feature extraction. It also facilitates automated generation of spectrograms and provides visual representations, preprocessed audio samples, as well as detailed data results.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003ePopulation dataset\u003c/h2\u003e \u003cp\u003eA total of 84 patients were recruited, including 43male individuals with an average age of (65.50\u0026thifnsp;\u0026plusmn;\u0026thinsp;9.45) years and 41 female individuals with an average age of (64.90\u0026thinsp;\u0026plusmn;\u0026thinsp;6.48) years. The details of demographics of the subjects refer the Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDemographics of subjects.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003cp\u003e(mean\u0026thinsp;\u0026plusmn;\u0026thinsp;std)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003cp\u003e(mean\u0026thinsp;\u0026plusmn;\u0026thinsp;std)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeart rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71.60\u0026thinsp;\u0026plusmn;\u0026thinsp;4.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68.80\u0026thinsp;\u0026plusmn;\u0026thinsp;6.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlood oxygen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e96.38\u0026thinsp;\u0026plusmn;\u0026thinsp;1.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e98.00\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65.50\u0026thinsp;\u0026plusmn;\u0026thinsp;9.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e64.90\u0026thinsp;\u0026plusmn;\u0026thinsp;6.48\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e176.12\u0026thinsp;\u0026plusmn;\u0026thinsp;5.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e159.78\u0026thinsp;\u0026plusmn;\u0026thinsp;4.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e69.32\u0026thinsp;\u0026plusmn;\u0026thinsp;9.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e50.78\u0026thinsp;\u0026plusmn;\u0026thinsp;6.12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22.38\u0026thinsp;\u0026plusmn;\u0026thinsp;3.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e19.92\u0026thinsp;\u0026plusmn;\u0026thinsp;2.50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eVisualization and denoise the respiratory sound\u003c/h2\u003e \u003cp\u003eBefore standardization, the collected respiratory sounds were difficult to identify due to interference from heart sounds in the three auscultation areas on the left side, which were close to the heart (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In contrast, the three auscultation areas on the right side experienced less interference from heart sounds (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). After undergoing standardization processing, the characteristics of respiratory sounds became evident, resulting in noise reduction and signal amplification effects, leading to clearer waveforms (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eRespiratory sound cycle segmentation\u003c/h2\u003e \u003cp\u003eThe segmentation of multiple respiratory cycles from an audio file was based on the localization of two adjacent \"similar images\" in phonocardiography (PCG)(defined as two consecutive images with similar amplitude and duration) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eThe inhalation phase (S1) and exhalation phase (S2) separation\u003c/h2\u003e \u003cp\u003eAfter positioning and segmenting the respiratory cycle, audio data analysis tools can identified the intensity, duration, and frequency domain characteristics of S1 and S2 as follows: (I) S1 has a longer duration than S2 and its amplitude was significantly higher than that of S2. (II) Both S1 and S2 have a certain frequency range in physiological terms. (III) In PCG, S2 occurs between adjacent S1 signals. By combining these features, we determined the start and end positions of both S1 and S2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eSegmentation of inhalation and exhalation phases based on label information.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe time tag information was extracted from the Respirtory Data Analysis Tool to accurately separate the inhalation and exhalation periods. After identifying and locating S1 and S2, label values were calculated (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003e) to precisely clip the recordings. The recorded respiratory sounds include precise start and end times for each breathing cycle.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe amplitude of respiratory sounds can indicate the level of physical activity in the respiratory muscles, and previous studies suggested that this could be valuable for COVID-19 detection [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Sound magnitude was utilized to calculate and filter loudness values, specifically determining the amplitudes of different segments (S1, S2) within the respiratory cycle. In this study, we computed the average and standard deviation of amplitude features (Maxdb, Mindb, Meandb, Midledb) for each segment of the respiratory cycle.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eSpectrum\u003c/h2\u003e \u003cp\u003eSpectrum analysis is the most widely used method in respiratory sound analysis. In this study, MATLAB code (Supplementary data) was used to perform formula calculations on two segments (S1 and S2).\u003c/p\u003e \u003cp\u003e \u003cb\u003eCross-validation based on the respiratory cycle.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIn order to validate the accuracy of our segmentation results, the double-blind validation also conducted respiratory phase verification with experienced experts in the field of respiratory diseases. This was performed to record the subjects' respiratory rates during the data collection process. \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eRespiratory system diseases are one of the most common health issues affecting people in their daily lives. With a high incidence rate and limited diagnostic methods, there is an urgent need for simple, reliable, and radiation-free diagnostic techniques. Intelligent auscultation technology for respiratory sounds can play a significant role in predicting respiratory system diseases [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, during the performing respiratory auscultation in a clinical frontline, even experienced respiratory specialists can be disturbed in the diagnosis of respiratory disorders due to the potential influence of a variety of external factors, which may lead to misdiagnosis or omission.\u003c/p\u003e \u003cp\u003eThe artificial intelligence recognition technology is capable of detecting acoustic information from respiratory sound signals, encompassing temporal domain features (e.g., energy, average value, peak value), frequency domain features (e.g., spectral energy, frequency peak), and time-frequency domain features (e.g., wavelet coefficients). Moreover, this method presented the advantages of batch processing and convenience in screening and diagnosing respiratory system diseases [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].In this study, we developed and validated an automated method for segmenting respiratory sounds and extracted basic acoustic information and sound intensity parameters using our audio data analysis tool. Through data analysis, no significant differences were found in respiratory sound against the respiratory system diseases between detected by the audio data analysis tool and radiological examinations.\u003c/p\u003e \u003cp\u003eDiagnosis of respiratory diseases can be greatly improved by utilizing intelligent digital auscultation technology. McLane et al. presented a comprehensive signal analysis method for analyzing challenging recordings of crackles, which includes five processing modules: (1) motion artifact detection, (2) deep learning denoising network, (3) respiratory cycle segmentation, (4) separation of discontinuous non-tonal components from bubble sounds, and (5) crackle peak detection. These methods provide quantifiable analysis of respiratory sounds, enabling clinicians to differentiate various types of crackles based on their timing and severity within the respiratory cycle [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].To address the occasional misidentification of respiratory sounds by medical professionals, Kim et al. developed an automatic classification system for respiratory sounds by using of deep learning convolutional neural networks (CNN). They successfully categorized 1918 types of respiratory sounds from clinical records into various categories including normal, crackles, wheezes, and rhonchi. By integrating a pre-trained sequential image feature extractor with the respiratory sound data and CNN classifier, they established a predictive model for accurately classifying respiratory sounds. This innovative model holds great potential in facilitating prompt diagnosis and appropriate treatment of respiratory system diseases [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].Heitmann et al. used 35.9 hours of auscultation audio data from 572 pediatric outpatient patients to develop Deep Breath: a deep learning model for identifying sound features of acute respiratory diseases in children. The model consists of a convolutional neural network and a logistic regression classifier, which aggregated estimates recorded from eight chest locations into a single prediction at the patient level. Internal 5-fold cross-validation was performed on the results, and external validation was conducted in three other countries (Senegal, Cameroon, Morocco). Deep Breath provides an interpretable framework for deep learning to identify objective audio features of respiratory diseases [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].The major of existing lung sound recognition methods neglect the correlation between temporal and spectral information of lung sounds, resulting in insufficient capturing of detailed features. Shi et al. introduced a model that incorporates wavelet feature enhancement and time-frequency synchronized modeling techniques. This model comprises a dual wavelet analysis module (DWAM), a cubic network, and an attention module. Experimental results on both merged datasets as well as those from the 2017 International Conference on Biomedical and Health Informatics demonstrate that their proposed framework outperforms existing models by over 1.36% and 4.28%, respectively [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e Our research shares similarities with the theirs, commencing from the rigorous selection of participants based on specific inclusion and exclusion criteria, employing advanced electronic stethoscopes for capturing respiratory sound data, followed by preprocessing of sound signals, extraction of relevant features, and ultimately achieving comprehensive analysis of respiratory sounds. However, it has been suggested in certain studies that harnessing machine learning models for analyzing lung sounds can further explore their diagnostic potential for other diseases related to the respiratory system [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe innovation in our respiratory sound signal processing embedded in the utilization of our self-developed software, Repertory Data Analysis Tool, for data processing. This tool effectively aggregates time labels for respiratory sounds and employs MATLAB for audio analysis, feature extraction, and the generation of partial spectrograms. In addition, all acoustic parameters can be collected, which also facilitates researchers to diversify the data.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe integration of electronic stethoscope and artificial intelligence technology has facilitated the digital auscultation technique, enabling the digitized collection of respiratory sounds and precise identification of inspiratory phase (S1) and expiratory phase (S2). The analysis results obtained through artificial intelligence technology on respiratory sounds demonstrate robust scientific stability. This approach provides objective evidence for respiratory auscultation and noise assessment, thus holding significant importance in clinical research.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eLimitations\u003c/strong\u003e \u003cp\u003eDespite our efforts to minimize noise during data collection, significant amounts of noise were present due to uncontrollable factors such as heart sounds, poor stethoscope fit on thin patients, and external environmental factors. Additionally, manually cutting audio files at their beginnings and endings during sound processing could potentially affect data analysis results.\u003c/p\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe ethics committee at Ethics Committee of Shanghai Changhai Hospital approved the investigation (CHEC2024-049). The written informed consent has been obtained from all participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author will provide the original data supporting the conclusions of this article without any unnecessary withholding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author(s) declare that financial support was received for the research, authorship, and/or publication of this article.\u0026nbsp;This work was supported by Special Fund for \u0026ldquo;\u003cstrong\u003e\u003cem\u003eResearch on Community Medicine and Health Management\u0026rdquo; in Shanghai (2023SQ01). Shanghai \u0026quot;Rising Stars of Medical Talent\u0026quot; Youth Development Program\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e（\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003eYouth Medical Talents\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e\u0026ndash;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e\u0026nbsp;General Practitioner Program\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e）\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFJ, NHR, CXL: Data collection management,\u0026nbsp;Formal analysis, Resources, Ethical application, Writing original draft. DYL: Formal analysis, Validation, Project management, Resources, Writing original draft. WWM : Data collection, Data classification management. XF, SY: Methodology, Grant acquisition, Project management, Review and editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMcLane I, Lauwers E, Stas T, et al. Comprehensive Analysis System for Automated Respiratory Cycle Segmentation and Crackle Peak Detection. IEEE J Biomed Health Inform 2022; 26(4): 1847\u0026ndash;1860\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu W, He G, Pan C, et al. A forced cough sound based pulmonary function assessment method by using machine learning. Front Public Health 2022; 10: 1015876\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaj V, Renjini A, Swapna MS, Sreejyothi S, Sankararaman S. Nonlinear time series and principal component analyses: Potential diagnostic tools for COVID-19 auscultation. Chaos Solitons Fractals 2020; 140: 110246\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChao HS, Tsai CY, Chou CW, et al. Artificial Intelligence Assisted Computational Tomographic Detection of Lung Nodules for Prognostic Cancer Examination: A Large-Scale Clinical Trial. Biomedicines 2023; 11(1): 147\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQi J, Hong B, Tao R, et al. Prediction model for malignant pulmonary nodules based on cfMeDIP-seq and machine learning. Cancer Sci 2021; 112(9): 3918\u0026ndash;3923\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen M, Copley SJ, Viola P, Lu H, Aboagye EO. Radiomics and artificial intelligence for precision medicine in lung cancer treatment. Semin Cancer Biol 2023; 93: 97\u0026ndash;113\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang DM, Huang J, Qiao K, Zhong NS, Lu HZ, Wang WJ. Deep learning-based lung sound analysis for intelligent stethoscope. Mil Med Res 2023; 10(1): 44\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD Folgado MB, M Antunes MLN, Liu H. Tssearch: Time series subsequence search library. SoftwareX 2022;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZF Bertrand KDS, S\u0026aacute;nchez DI. Lung auscultation in the 21th century. Revista Chilena de \u0026hellip;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eV Raj MSS, Sankararaman S. Bioacoustic signal analysis through complex network features. Comput Biol Med 2022;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Z, Li M, Wang R, et al. Diagnosis of COVID-19 via acoustic analysis and artificial intelligence by monitoring breath sounds on smartphones. J Biomed Inform 2022; 130: 104078\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee SH, Kim YS, Yeo MK, et al. Fully portable continuous real-time auscultation with a soft wearable stethoscope designed for automated disease diagnosis. Sci Adv 2022; 8(21): eabo5867\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahman T, Ibtehaz N, Khandakar A, et al. QUCoughScope: An Intelligent Application to Detect COVID-19 Patients Using Cough and Breath Sounds. Diagnostics (Basel). 2022;12(4):920. Published 2022 Apr 7. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics12040920\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics12040920\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGupta P, Moghimi MJ, Jeong Y, Gupta D, Inan OT, Ayazi F. Precision wearable accelerometer contact microphones for longitudinal monitoring of mechano-acoustic cardiopulmonary signals. NPJ Digit Med 2020; 3: 19\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim Y, Hyon Y, Jung SS, et al. Respiratory sound classification for crackles, wheezes, and rhonchi in the clinical field using deep learning. Sci Rep 2021; 11(1): 17186\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeitmann J, Glangetas A, Doenz J, 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\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi L, Zhang Y, Zhang J. Lung Sound Recognition Method Based on Wavelet Feature Enhancement and Time-Frequency Synchronous Modeling. IEEE J Biomed Health Inform 2023; 27(1): 308\u0026ndash;318\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDar JA, Srivastava KK, Mishra A. Lung anomaly detection from respiratory sound database (sound signals). Comput Biol Med 2023; 164: 107311\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeitmann J, Glangetas A, Doenz J, 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. Published 2023 Jun 2. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41746-023-00838-3\u003c/span\u003e\u003cspan address=\"10.1038/s41746-023-00838-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi L, Zhang Y, Zhang J. Lung Sound Recognition Method Based on Wavelet Feature Enhancement and Time-Frequency Synchronous Modeling. IEEE J Biomed Health Inform. 2023;27(1):308\u0026ndash;318. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/JBHI.2022.3210996\u003c/span\u003e\u003cspan address=\"10.1109/JBHI.2022.3210996\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDar JA, Srivastava KK, Mishra A. Lung anomaly detection from respiratory sound database (sound signals). Comput Biol Med. 2023;164:107311. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2023.107311\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2023.107311\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Respiratory sound, Artificial intelligence, Digital auscultation, Automatic segmentation method, Respiratory system diseases","lastPublishedDoi":"10.21203/rs.3.rs-5324173/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5324173/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e \u003cb\u003eBackground\u003c/b\u003e:In the field of respiratory system diseases, the utilization of respiratory sounds in auscultation plays a crucial role in the specific disease diagnosis. However, during the process of auscultation, the personal experiences and environmental factors may affect the decision making, leading to diagnostic errors. Therefore, to accurately and effectively obtaining and analyzing respiratory sounds can be positively contribute to the diagnosis and treatment of respiratory system diseases.\u003c/p\u003e \u003cp\u003e \u003cb\u003eObjectives\u003c/b\u003e:Our aim was to develop an analytical method for the visualization and digitization of respiratory audio data, and to validate its capability to differentiate between various background diseases.\u003c/p\u003e \u003cp\u003e \u003cb\u003eMethods\u003c/b\u003e:This study collected the respiratory sounds of patients admitted to the Department of General Medicine of Shanghai Changhai Hospital from June to December 2023. After strict screening according to the inclusion and exclusion criteria, a total of 84 patients were included. The research process includes using an electronic stethoscope to collect lung sounds from patients in a quiet environment. The patients expose their chests and lie flat. Sound data is collected at six landmark positions on the chest. The collected audio files are imported into an analysis tool for segmentation and feature extraction. Specific analysis methods include distinguishing heart sounds and respiratory sounds, segmenting respiratory sounds, determining the inspiratory and expiratory phases, and using a tool developed by the team for automatic segmentation encoding.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResults\u003c/b\u003e:We standardized the respiratory sounds of 84 patients and segmented multiple respiratory cycles. Following the localization and segmentation of the respiratory cycles based on label information, we calculated the average and standard deviations of the amplitude features for each segment of the respiratory cycle. The results indicated differences among various diseases.\u003c/p\u003e \u003cp\u003e \u003cb\u003eConclusions\u003c/b\u003e:The robust algorithm platform is capable to segmenting the respiratory sounds into inhale and exhale phase accordingly, then to comparing the difference between different background disease. This method provides objective evidence for auscultation of respiratory sounds and visual display of breath sounds.\u003c/p\u003e","manuscriptTitle":"Construction and validation of an automatic segmentation method for respiratory sound time labels","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-11 06:57:39","doi":"10.21203/rs.3.rs-5324173/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9bcfd97a-9010-4995-973e-05e54ec6674d","owner":[],"postedDate":"November 11th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-01-21T13:53:37+00:00","versionOfRecord":[],"versionCreatedAt":"2024-11-11 06:57:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5324173","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5324173","identity":"rs-5324173","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}