Breathing sounds analysis system for early detection of airway problems in patients with a tracheostomy tube

doi:10.21203/rs.3.rs-3233084/v1

Download PDF

Article

Breathing sounds analysis system for early detection of airway problems in patients with a tracheostomy tube

https://doi.org/10.21203/rs.3.rs-3233084/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 29 Nov, 2023

Read the published version in Scientific Reports →

You are reading this latest preprint version

To prevent immediate mortality in patients with a tracheostomy tube, it is essential to ensure timely suctioning or replacement of the tube. Breathing sounds at the entrance of tracheostomy tubes were recorded with a microphone and analyzed using a spectrogram to detect airway problems. The sounds were classified into three categories based on the waveform of the spectrogram according to the obstacle status: normal breathing sounds (NS), vibrant breathing sounds (VS) caused by movable obstacles, and sharp breathing sounds (SS) caused by fixed obstacles. A total of 3,950 breathing sounds from 23 patients were analyzed. Despite neither the patients nor the medical staff recognizing any airway problems, the number and percentage of NS, VS, and SS were 1,449 (36.7%), 1,313 (33.2%), and 1,188 (30.1%), respectively. Artificial intelligence (AI) was utilized to automatically classify breathing sounds. MobileNet and Inception_v3 exhibited the highest sensitivity and specificity scores of 0.9441 and 0.9414, respectively. When classifying into three categories, ResNet_50 showed the highest accuracy of 0.9027, and AlexNet showed the highest accuracy of 0.9660 in abnormal sounds. Classifying breathing sounds into three categories is very useful to decide whether to suction or change the tracheostomy tubes, and AI can accomplish this with high accuracy.

Biological sciences/Biotechnology

Health sciences/Health care

Health sciences/Medical research

Tracheostomy

Respiratory sound

Breathing sound

Artificial intelligence

Deep learning

Machine learning

As the population gets older, the number of tracheostomies has been increasing. Parker et al. reported that respiratory failure, ineffective cough, neurological injury, and carcinoma were increasing in older people and driving the increase in the number of tracheostomies ¹. Performing a tracheostomy requires utmost attention, and meticulous care thereafter. Patients with tracheostomies often produce more sputum than healthy individuals due to inadequate air filtering and humidification during breathing. If the tube is blocked by any obstacle partially or completely, it can cause serious problems for patients, especially fixed obstacles, which can block the tube completely. To prevent this, humidification must be supplied. Because dehydration hinders the expectoration of sticky secretions, humidification is important. Next, proper suction is needed. Despite the humidity, if sputum is not removed and phlegm accumulates, it can block the tube. However, suctioning too frequently is also undesirable because the suction stimulates the trachea wall, which causes contractions and can lead to hypoxia ¹. For these reasons, sputum removal needs to be performed gently at an adequate time. However, it is one of the difficult skills to learn and has the disadvantage of being subjective. There are barriers to entry that are challenging for ordinary people, including inexperienced medical staff.

Artificial intelligence (AI) is being widely incorporated into healthcare. Many symptoms and diseases are now analyzed and even diagnosed by AI, mainly using images ^2,3, but some trials have also used sounds ^4–7. Shi et al. ⁴ reported that they used AI to classify sputum sounds from patients with intubation tubes. Srivastava et al. ⁵ reported that AI helped detect chronic obstructive pulmonary disease using respiratory sounds. Nakano et al. ⁷ reported that AI helped to detect sleep apnea through tracheal sounds.

Comprehensively, we hypothesized that AI could be used to analyze the breathing sounds of tracheostomy patients. Our goal in this study was to classify the breathing sounds more subdivided and attempt to classify them through AI.

1. Patients

This prospective study was performed at a single university center from March 2021 to April 2022 among patients who received an elective tracheostomy. Patients who needed ventilator unit care after tracheostomy or underwent an emergency tracheostomy in the emergency room or intensive care unit (ICU), patients younger than 20 years, and pregnant patients were excluded. Using those criteria, 23 patients with tracheostomy were enrolled in this study. We obtained the following clinical information for all patients. The study was approved by Bucheon St. Mary Hospital of the Catholic University of Korea institutional review board (IRB) (The physiologic changes of trachea according to the degree of sputum after tracheostomy, HC20ONSI0106, approved November 10, 2020). Procedures were followed in accordance with IRB ethical standards and with the Helshinki Declaration of 1975.

2. Recording system

Breathing sound samples were recorded with a voice recorder (Model PCM-A10; Sony, Japan) using a condenser microphone (Model ECM-CS3; Sony, Japan) located two or three cm from the outer opening of the tracheostomy tube in line with the direction of the tube. The recording type was linear pulse code modulation, which can record original sounds without compression.

3. Data collection and classification

All data collection started in the ICU immediately after surgery. Patients were transferred to the general ward two or three days after tracheostomy, and the existing tracheostomy tube was replaced with a new fenestration-type tracheostomy tube three to five days after surgery. We collected data only until the tube was changed. As a result, all patients were recorded for an average of 12 to 16 hours a day for 1 to 5 days after tracheostomy.

Breathing sounds with severe background noise and very low breathing sounds that were not detected in the spectrogram due to some reasons such as sleep were excluded. Breathing sounds during the period when the patients expressed severe dyspnea or medical staff observed airway problems were also excluded. Although patients sometimes noticed the presence of sputum coming out of the tracheostomy tube, there were no other abnormal symptoms such as dyspnea in the finally included data. Two expert otorhinolaryngology doctors with more than 10 years of experience heard all the recording samples and analyzed the waveform of the spectrogram. All the breathing sounds were classified into three categories: normal breathing sound (NS); low-frequency vibrant breathing sound (VS) that indicates a movable obstacle such as sputum in the tracheostomy tube that requires suctioning; high-frequency sharp breathing sound (SS) that indicates a fixed obstacle, including crusts, and blood clots in the tracheostomy tube that requires suctioning or changing the inner cannula of the tube. If the results between expert opinions and the waveform of the spectrogram are different, the result of the spectrogram was selected. If the sound was both vibrant and sharp, it was classified as VS.

All patients also wore parts of a polysomnography device during the study. The oronasal airflow device was located at the opening of the tracheostomy tube and did not interfere with breathing or recording. However, electroencephalogram, electrooculogram, and electromyography sensors were not set up. A photograph of the devices installed on each patient is shown in Fig. 1.

4. Breathing sound data processing methods

In this study, we converted breathing sound samples into a spectrogram and Mel frequency cepstral coefficient (MFCC), which are audio features widely utilized to analyze respiratory status. The details of the converted features are described in the following sections. All data processing for sound classification by multiple AI algorithms, such as spectrogram conversion and MFCC extraction, was performed in MATLAB 2019a.

4 − 1. Spectrogram conversion

An audio spectrogram is a two-dimensional image that simultaneously presents sound waveforms and spectra. By representing continuously changing spectra as a data sample, spectrograms provide rich audio information and are widely used in deep learning frameworks based on image classification ^7–9. The breathing sound samples were converted into spectrograms using a short-time Fourier transform. A more detailed process of conversion is presented in Supplementary 1.

Examples of the time-domain wave characteristics and spectrograms of breathing sounds are shown in Fig. 2. In NS, because the airway had no obstacles and there was almost no friction, the acoustic energy was relatively small and mainly below 2,000 Hz. On the other hand, the acoustic energy was scattered over a large area, from 500 to 12,000 Hz, in the abnormal breathing sounds because the obstacles generated sounds of various frequencies. VS was a repetitive pattern at a frequency of less than 100 Hz that occurred when a movable obstacle blocked the trachea or tracheostomy tube. This pattern appears as multiple vertical lines in the spectrogram. In contrast, SS occurred when stiff or fixed obstacles narrowed the cross-section of the airway, which induced a wide range of high-frequency breathing sounds that were more continuous than VS. The pattern of SS appears as multiple horizontal lines in the spectrogram.

4 − 2 MFCC extraction

The MFCC is a group of audio parameters suitable for human auditory characteristics and has been widely applied for speech recognition ^10,11 and respiratory diagnosis ^12–14. In this study, we used MFCC as a tracheal breathing sound feature for machine learning–based classifiers. A more detailed process of extraction and the designed filter banks are presented in Supplementary 2 and 3.

5. Breathing sound classification methods

5 − 1 MFCC-based Machine learning classification methods

For MFCC-based breathing sound classification, a support vector machine (SVM) ^15,16 and k-nearest neighbor (kNN) ^17,18, which are widely used for health status diagnosis using MFCC, are employed. And all breathing sound classification for the machine learning algorithms was performed using a desktop machine with an Intel i5-10500F CPU and NVIDIA GeForce RTX1660 Ti (6 GB) GPU. A more detailed process of extraction is presented in Supplementary 4.

5 − 2. Spectrogram-based deep learning method: Convolutional neural network (CNN)

CNN is a widely and successfully established deep learning algorithm in the field of image classification and pattern recognition ¹⁹. The basic structure of a CNN is the convolution layer, which calculates the tensor transferred to the next layer through a convolution computation between the tensor and the kernel. A CNN topology includes many convolution layers designed based on factors such as kernel size and number. A CNN provides a framework for learning the features common among images in a data group without requiring manual data extraction, and it can generate accurate pattern recognition or image classification–trained models. Biomedical signal classification studies have been accelerated by imaging and learning with CNNs ^7–9. For example, CNN classification based on spectrograms has increased the accuracy of respiratory pattern classification in many clinical fields. In this study, we applied spectrograms converted from one cycle of respiratory data to the following CNN topologies: AlexNet ²⁰, VGGNet ²¹, ResNet ²², Inception_v3 ²³, and MobileNet ²⁴. The CNN classification was conducted with Lenovo Intelligent-Computing-Orchestration with a batch size of 32 and a maximum iteration of 200,000.

1. Patients

We enrolled and analyzed 23 patients in this study, of whom 21 were male. The average age of all patients was 66 years. Six patients were never smokers, and seven patients were heavy smokers with a history of 30 pack years or more. No patient had a history of asthma, but one patient had been treated for tuberculosis prior. Forced vital capacity and forced expiratory volume at 1 second were measured in 15 patients. Four patients were diagnosed with chronic obstructive pulmonary disease, one patient was diagnosed with restrictive pulmonary disease, and one patient was diagnosed with mixed-type pulmonary disease. Detailed patient information is provided in Table 1.

Table 1

Clinical characteristics of the 23 participants with a tracheostomy.
Characteristics		Number of patients (%) or mean (minimum, maximum)
Sex	Male	21 (91.3)
	Female	2 (8.7)
Age (years)		66 (53,79)
Height (cm)		163 (152,175)
Weight (kg)		65 (52,81)
Cause of disease	Cancer	21 (91.3)
	Goiter	1 (4.3)
	Subglottic stenosis	1 (4.3)
Smoking	Never	6 (26.1)
	≤ 30 pyrs	10 (43.5)
	> 30 pyrs	7 (30.4)
Hypertension		10 (43.5)
Diabetes		2 (8.7)
Cardiovascular disease		2 (8.7)
Asthma		0 (0)
Tuberculosis treatment history		2 (8.7)
Previous operation history in head and neck region		2 (8.7)
Radiotherapy history		1 (4.3)
FEV1/FVC	≥ 70%	10 (43.5)
	≥ 60% and < 70%	4 (17.4)
	< 60%	1 (4.3)
	No data	8 (34.8)
FVC/Predicted FVC	≥ 80%	13 (56.5)
	< 80%	2 (8.7)
	No data	8 (34.8)
Pulmonary disease	Normal	9 (39.2)
	Restrictive	1 (4.3)
	Chronic obstructive	4 (17.4)
	Mixed	1 (4.3)
	No data	8 (34.8)
Abbreviations:
pyrs, pack years; FVC, forced vital capacity; FEV1, forced expiratory volume at 1 second

2. Analysis of the breathing sounds

A total of 3,950 breathing sounds from 23 patients were analyzed by segmenting the respiratory cycles in the recording samples. Although both the patients and the medical staff next to them judged that there was no problem in the airway of the patients during the collection of breathing sounds, the breathing sounds were classified into three categories after analysis using a microphone-based recording system and the waveform of the spectrogram. The number and percentage of NS, VS, and SS samples were 1,449 (36.7%), 1,313 (33.2%), and 1,188 (30.1%), respectively. The mean time of each respiration cycle was 3.801 ± 1.592, 3.033 ± 0.949, and 3.876 ± 1.246 seconds, respectively. Detailed information for each breathing sound is given in Table 2.

Table 2

**Characteristics of the 3,950 breathing sounds from 23 patients.** All breathing sounds were segmented by respiratory cycle and classified into three categories: normal breathing sounds (NS), vibrant breathing sounds (VS), and sharp breathing sounds (SS).
	Normal sound (NS)	Vibrant sound (VS)	Sharp sound (SS)	Total
Number of breathing sounds (%)	1,449 (36.7)	1,313 (33.2)	1,188 (30.1)	3,950 (100)
Number per person (mean ± SD)	63 ± 38.8	57.1 ± 57.0	51.6 ± 43.9	171.7 ± 75.0
Time per 1 cycle of respiration (mean ± SD) (sec)	3.801 ± 1.592	3.033 ± 0.949	3.876 ± 1.246	3.484 ± 1.292

3. Accuracy of automatic detection of breathing sound samples

Of the 3,950 breathing sounds, 3,150 (80%) were used as training data, and 791 (20%) were used as testing data. First, we evaluated the accuracy, sensitivity, specificity, positive predictive value, and negative predictive value using binary classification: normal and abnormal breathing sounds. MobileNet and Inception_v3 showed the highest values in sensitivity and specificity, with scores of 0.9441 and 0.9414, respectively. ResNet_50 achieved the highest accuracy of 0.9330 and exhibited excellent performance overall across sensitivity, specificity, positive predictive value, negative predictive value, and area under the curve (Table 3). The receiver operating characteristic curves of each classifier are plotted in Fig. 3. When we used three categories, we measured the total accuracy and the “accuracy in abnormal sounds,” which is the accuracy in distinguishing between VS and SS. Among the CNN algorithms, ResNet_50 showed the highest accuracy of 0.9027, and AlexNet showed the highest accuracy of 0.9660 in abnormal sounds (Table 4). Comprehensively, the CNNs showed better classification performances than the machine learning (ML) models (SVM and kNN).

Table 3

Evaluation metrics table to classify breathing sounds of 23 patients with a tracheostomy tube: normal and abnormal sounds
	Input	Name of the classifier	Accuracy	Sensitivity	Specificity	Positive predictive value	Negative predictive value	Area under curve
CNN	Spectrogram	AlexNet	0.9140	0.9401	0.8690	0.9253	0.8936	0.9586
		VGG_16	0.9090	0.9341	0.8655	0.9231	0.8838	0.9527
		ResNet_50	0.9330	0.9381	0.9241	0.9553	0.8963	0.9650
		Inception_v3	0.9317	0.9261	0.9414	0.9647	0.8806	0.9608
		MobileNet	0.9241	0.9441	0.8897	0.9366	0.9021	0.9549
SVM	MFCC(20)	3rd polynomial	0.8306	0.8922	0.7241	0.8482	0.7955	0.9248
	MFCC(20)	RBF	0.9153	0.9142	0.9172	0.9502	0.8608	0.9681
	MFCC(40)	3rd polynomial	0.8609	0.8703	0.8448	0.9064	0.7903	0.9257
	MFCC(40)	RBF	0.9102	0.9142	0.9034	0.9424	0.8590	0.9598
kNN	MFCC(20)	k = 3	0.8900	0.9142	0.8483	0.9124	0.8512	0.9098
		k = 5	0.8786	0.8882	0.8621	0.9175	0.8170	0.9159
		k = 7	0.8824	0.8842	0.8793	0.9268	0.8147	0.9369
	MFCC(40)	k = 3	0.9102	0.9341	0.8690	0.9249	0.8842	0.9102
		k = 5	0.8976	0.9102	0.8759	0.9268	0.8495	0.9249
		k = 7	0.8913	0.9042	0.8690	0.9226	0.8400	0.9406
Abbreviations: CNN, convolution neural network; SVM, support vector machine; kNN, k-nearest neighbor; RBF, radial bias function

Table 4

**Evaluation metrics table to classify the breathing sounds of 23 patients with a tracheostomy tube into three categories: NS, VS, and SS.** We also measured “accuracy in abnormal sounds” to distinguish between VS and SS among the abnormal breathing sounds.
	Input	Name of the classifier	Accuracy	Accuracy in abnormal sounds
CNN	Spectrogram	AlexNet	0.8938	0.9660
		VGG_16	0.8799	0.9509
		ResNet_50	0.9027	0.9489
		Inception_v3	0.9001	0.9461
		MobileNet	0.8951	0.9514
SVM	MFCC(20)	3rd polynomial	0.7295	0.8210
	MFCC(20)	RBF	0.8217	0.8384
	MFCC(40)	3rd polynomial	0.6738	0.6606
	MFCC(40)	RBF	0.8306	0.8625
kNN	MFCC(20)	k = 3	0.8458	0.9236
		k = 5	0.8319	0.9169
		k = 7	0.8268	0.9007
	MFCC(40)	k = 3	0.8711	0.9338
		k = 5	0.8496	0.9167
		k = 7	0.8331	0.8985
Abbreviations: NS, normal breathing sounds; VS, vibrant breathing sounds; SS, sharp breathing sounds; CNN, convolutional neural network; SVM, support vector machine; kNN, k-nearest neighbor; RBF, radial bias function

There are two major ways to secure an airway. One is intubation, and the other is tracheostomy. After either of those procedures, management, and care of the tube are important. In most cases of intubation, a ventilator, which controls humidity and alarms the status of the tube based on waveforms of respiratory pressure, is supplied ²⁵. In cases of tracheostomy, on the other hand, many patients maintain spontaneous respiration, and the judgment of obstruction is done by the patient himself or nearby medical staff. However, the judgment of the patient is not always accurate. And medical staff has many other important works to do, so they cannot concentrate only on the respiration of the patient. In addition, this must be done by experienced medical staff. Even in a complex and noisy environment, an objective diagnostic tool on the same level as an experienced medical staff focusing only on respiration is needed, and AI can be considered.

Lung sounds are usually classified into two groups: normal and abnormal or adventitious ⁶. However, there is no well-known classification for sputum sound. Some reports tried to classify them, but they succeeded only in detecting normal and abnormal breathing sounds ^4,26. We decided to classify the sounds into three categories based on the status of the obstacles, proved by the waveform of the spectrogram: NS, VS, and SS. This difference was expressed not only as the sounds but also as the waveform of the spectrogram. The sound energy of NS was concentrated below 2,000 Hz, whereas the sound energy of SS was distributed above 4,000 Hz. VS, which involves repeated sound patterns for a very short time, showed the characteristic vertical lines. (Fig. 1) These characteristic spectrogram waveforms are the visual basis for the three categories` classification according to the status of the obstacles. From a clinical point of view, if VS is heard in the tracheostomy tube, most cases can be improved with only suction. However, if SS is heard in the tracheostomy tube, some cases might need the change of tube or inner cannula. Of course, in some cases, both movable and fixed obstacles can be present in the airway, and both VS and SS will be heard. We classified those occasions as VS because such events might be improved with only suction, and suction is safer and faster than a tube change.

Direct listening was affected by environmental noise. In this study, since the recording was conducted in a noisy environment, such as the ICU, no medical staff next to the patients recognized the status of the tube. However, as a result of using a recording system with a microphone, many recordings among them were judged as VS and SS. Even if an omnidirectional microphone was used, the microphone was positioned facing the tube tip and as close as possible because the direction and distance of the microphone were important factors. If a microphone and tracheostomy tubes are developed as an integrated unit, it is expected that the sensitivity and accuracy of the classification system will further increase.

When we compared the CNNs with the ML, the CNNs showed much higher accuracy. Since NS and SS have different focused frequency ranges, both CNN and ML could distinguish NS and SS well. However, VS was much better discriminated against by CNN than by ML. Whereas ML judged the type of breathing sounds based on an entire cycle of respiration, CNN could discriminate a temporal element because it analyzed data that had been converted into a spectrogram.

We used polysomnography to assess oxygen saturation, heart rate, blood pressure, respiratory rate, and airflow. We thought that we might find significant correlations between airway obstruction and aerodynamic factors, but we did not, presumably because the aerodynamic factors were controlled more by factors such as pain and sleep than by the degree of airway obstruction. In cases of severe airway obstruction, some aerodynamic factors showed changes, such as respiratory rate and airflow pressure, but those changes were inconsistent.

This study has some limitations. First, these breathing sounds were not obtained in a soundproof facility. Although we excluded sounds that contained severe noise, all our sound data contained some noise. Second, only two of the participants were female. Of course, the breathing sounds from the tracheostomy tube were not much affected by sex differences because the tracheostomy tube was located under the vocal folds and did not process the sound. Nevertheless, the next study should include more female patients.

Despite those limitations, this study is meaningful. First, this is the first study of breathing sounds from a tracheostomy tube. Second, this is the first trial to classify sputum sounds into three categories, confirmed by the spectrogram. Third, CNN classified sputum sounds as the accuracy of our system was very good, better than 90%.

Classifying breathing sounds into three categories instead of just normal and abnormal can be more useful in identifying several airway problems and determining the appropriate interventions even for caregivers who may not be expert medical staff. This system includes a recording system with a microphone and has an objective basis on the waveform of the spectrogram. If CNN is used, higher accuracy could be acquired.

Ethical approval and consent to participate

The study was approved by the institutional review board (IRB) of Bucheon St. Mary Hospital of the Catholic University of Korea. The approval number is HC20ONSI0106. All participants were provided with an explanation of the study and gave their informed consent. Consent documents were obtained from all participants.

Credit authorship contribution statement

HB Kim contributed to data curation, investigation, methodology, project administration, and writing original draft.

DY Koh contributed to formal analysis, investigation, methodology, software, and writing-original draft.

YH Jung and HJ Han contributed resources, visualization, and software.

JB Kim, and YH Joo contributed to conceptualization, funding acquisition, supervision, validation, and writing review & editing.

All authors have read and approved the final manuscript.

Data availability

The source codes used in this study are available online. The authors will also share sound data from deidentified individual participants in this study with researchers who provide a methodologically viable proposal and have the ability to analyze the data properly. Data-sharing requests can be directed to Younghoon Joo by email immediately following publication. To gain access to the data, requestors will need to sign a data access agreement.

Declaration of competing interest

The authors declare that they have no financial disclosures or conflicts of interest.

Acknowledgments

This work was supported by a Korean Medical Device Development Fund grant funded by the Korean government (Ministry of Science and ICT; Ministry of Trade, Industry, and Energy; Ministry of Health & Welfare, Ministry of Food and Drug Safety) (Project Number: RS-2020-KD000078, YH Joo)

Parker, V. et al. Trends and challenges in the management of tracheostomy in older people: the need for a multidisciplinary team approach. Contemp Nurse 26, 177-183 (2007).
Moen, E. et al. Deep learning for cellular image analysis. Nat Methods 16, 1233-1246 (2019).
Mazurowski, M. A., Buda, M., Saha, A. & Bashir, M. R. Deep learning in radiology: An overview of the concepts and a survey of the state of the art with focus on MRI. J Magn Reson Imaging 49, 939-954 (2019).
Shi, Y. et al. Classification of Sputum Sounds Using Artificial Neural Network and Wavelet Transform. Int J Biol Sci 14, 938-945 (2018).
Srivastava, A. et al. Deep learning based respiratory sound analysis for detection of chronic obstructive pulmonary disease. PeerJ Comput Sci 7, e369 (2021).
Pasterkamp, H., Kraman, S. S. & Wodicka, G. R. Respiratory sounds. Advances beyond the stethoscope. Am J Respir Crit Care Med 156, 974-987 (1997).
Nakano, H., Furukawa, T. & Tanigawa, T. Tracheal sound analysis using a deep neural network to detect sleep apnea. Journal of Clinical Sleep Medicine 15, 1125-1133 (2019).
Aykanat, M., Kılıç, Ö., Kurt, B. & Saryal, S. Classification of lung sounds using convolutional neural networks. EURASIP Journal on Image and Video Processing 2017, 1-9 (2017).
Perna, D. Convolutional neural networks learning from respiratory data in 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 2109-2113 (IEEE, 2018).
Tiwari, V. MFCC and its applications in speaker recognition. International journal on emerging technologies 1, 19-22 (2010).
Ittichaichareon, C., Suksri, S. & Yingthawornsuk, T. Speech recognition using MFCC in International conference on computer graphics, simulation and modeling (2012).
Meng, F., Shi, Y., Wang, N., Cai, M. & Luo, Z. Detection of respiratory sounds based on wavelet coefficients and machine learning. IEEE Access 8, 155710-155720 (2020).
Bahoura, M. & Ezzaidi, H. Hardware implementation of MFCC feature extraction for respiratory sounds analysis in 2013 8th International Workshop on Systems, Signal Processing and their Applications (WoSSPA) 226-229 (IEEE, 2013).
Bahoura, M. Pattern recognition methods applied to respiratory sounds classification into normal and wheeze classes. Computers in biology and medicine 39, 824-843 (2009).
Al-Angari, H. M. & Sahakian, A. V. Automated recognition of obstructive sleep apnea syndrome using support vector machine classifier. IEEE Transactions on Information Technology in Biomedicine 16, 463-468 (2012).
Palaniappan, R. & Sundaraj, K. Respiratory sound classification using cepstral features and support vector machine in 2013 IEEE Recent Advances in Intelligent Computational Systems (RAICS) 132-136 (IEEE, 2013).
Palaniappan, R., Sundaraj, K. & Sundaraj, S. A comparative study of the SVM and K-nn machine learning algorithms for the diagnosis of respiratory pathologies using pulmonary acoustic signals. BMC Bioinformatics 15, 223 (2014).
Haider, N. S., Singh, B. K., Periyasamy, R. & Behera, A. K. Respiratory Sound Based Classification of Chronic Obstructive Pulmonary Disease: a Risk Stratification Approach in Machine Learning Paradigm. J Med Syst 43, 255 (2019).
Gu, J. et al. Recent advances in convolutional neural networks. Pattern recognition 77, 354-377 (2018).
Hinton, G. E., Krizhevsky, A. & Sutskever, I. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems 25, 1 (2012).
Simonyan, K. & Zisserman, A. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014).
He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition in Proceedings of the IEEE conference on computer vision and pattern recognition 770-778 (2016).
Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J. & Wojna, Z. Rethinking the inception architecture for computer vision in Proceedings of the IEEE conference on computer vision and pattern recognition 2818-2826 (2016).
Howard, A. G. et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861 (2017).
Walter, J. M., Corbridge, T. C. & Singer, B. D. Invasive Mechanical Ventilation. South Med J 111, 746-753 (2018).
Niu, J. et al. Detection of sputum by interpreting the time-frequency distribution of respiratory sound signal using image processing techniques. Bioinformatics 34, 820-827 (2018).

No competing interests reported.

003supplementarytrachoAI0804plain.docx

Download PDF

Journal Publication

published 29 Nov, 2023

Read the published version in Scientific Reports →

Editorial decision: Major revision
15 Sep, 2023
Reviews received at journal
23 Aug, 2023
Reviewers agreed at journal
14 Aug, 2023
Reviewers agreed at journal
14 Aug, 2023
Reviewers invited by journal
13 Aug, 2023
Editor assigned by journal
09 Aug, 2023
Editor invited by journal
06 Aug, 2023
Submission checks completed at journal
06 Aug, 2023
First submitted to journal
03 Aug, 2023

You are reading this latest preprint version

Breathing sounds analysis system for early detection of airway problems in patients with a tracheostomy tube

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Methods

Results

Discussion

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1