Respiratory Disease Classification Lungs Sounds Using Machine Learning

Abstract

Globally, respiratory conditions like asthma, pneumonia, and chronic obstructive pulmonary disease (COPD) represent a serious threat to public health. Improvingpatientoutcomesandloweringhealthcarecosts depend on early and precise diagnosis. Despite being widely used, traditional stethoscope auscultation is constrainedbysubjectivityandinter-observervariability. This study investigates the use of machine learning methods to categorize respiratory conditions from recordings of lung sounds. To improve the quality of auscultation signals, preprocessing techniques like segmentation and noise filtering are used. To capture the temporalandspectralcharacteristicsofthelungsounds,a number of featuresareextracted,suchasspectralentropy, zero-crossing rate, and Mel Frequency Cepstral Coefficients (MFCCs). Support Vector Machines (SVM), Random Forests, and Convolutional Neural Networks (CNNs) are example of supervised learning algorithms that are trained.

Introduction

Background

Respiratory diseases (e.g., asthma, COPD, pneumonia) are major global health concerns, especially in low-resource areas where early diagnosis is challenging. Traditional diagnostic methods (like manual auscultation) are subjective and dependent on clinicians' expertise.

Emerging Solution

Advances in digital health and machine learning (ML) offer promising tools to automate respiratory disease diagnosis using lung sound recordings captured by electronic stethoscopes or microphones. These recordings contain vital acoustic patterns (wheezes, crackles, rhonchi) linked to various conditions.

Study Objective

Develop a machine learning framework for the automatic classification of respiratory diseases using lung sound data, focusing on:

• Feature extraction (e.g., MFCCs)

• Model training using ML/DL algorithms (SVM, Random Forest, CNN, RNN)

• Performance evaluation using public datasets like ICBHI 2017

Methodology

1. Data Collection

• Lung sound recordings labeled by medical experts.

• Datasets include conditions like asthma, COPD, pneumonia, and healthy controls.

2. Preprocessing

• Noise removal, segmentation, normalization, and resampling to improve data quality.

3. Feature Extraction

• Time, frequency, and time-frequency domain features (e.g., MFCCs, spectrograms).

4. Data Balancing & Augmentation

• SMOTE, oversampling, and audio augmentations (e.g., pitch shift) used to address class imbalance.

5. Model Training

• Traditional ML: SVM, Random Forest, k-NN

• Deep Learning: CNNs, RNNs (LSTM), Transformers

• Transfer learning used with pre-trained audio models (e.g., VGGish).

6. Evaluation

• Metrics: Accuracy, F1-score, AUC-ROC

• Tools: Confusion matrices, k-fold cross-validation

• Explainability: SHAP values, saliency maps

7. Deployment

• Flask-based web interface developed

• Models optimized for real-time, low-latency inference

• Ethical concerns (data privacy, fairness) addressed

• Multimodal integration (e.g., patient age, vitals) explored

Related Work

A review of past studies shows increasing effectiveness of:

• CNNs for pattern extraction (Demir & Sengur, 2020)

• RNNs (LSTM) for temporal dynamics (Perna & Tagarelli, 2019)

• Transformer models for multi-label classification (Kim & Kim, 2022)

• ML for COVID-19 diagnosis from cough/breath sounds (Pahar et al., 2021; Imran et al., 2020)

Results

• Validation Accuracy: Up to 95%

• F1-Score: Weighted average of 0.94

• COPD classification accuracy: 98.8%

• Some performance issues on underrepresented classes (e.g., URTI, bronchiolitis)

• No signs of overfitting; consistent training and validation curves

Additional Implementation Details:

• Model trained with Keras; best model saved via checkpointing.

• Web deployment using Flask

• Training logs show rapid improvement in accuracy over 70 epochs.

Conclusion

A promising,non-invasive method for the early detection and diagnosis of a variety of pulmonary conditions, including COPD, pneumonia, URTI, and bronchiectasis, is the respiratory disease classification system that uses lung sound analysis and machine learning. With validation accuracy peaking between 94 and 95%, the system achieves high accuracy by utilizing deep learning models trained on audio features extracted from lung sound recordings, especially in well-represented classes like COPD. Real-time audio processing, balanced training methods, and an intuitive web interface for clinical use all contribute to the model\'s performance. This method has great potential for clinical integration, providing scalable, easily accessible diagnostic supportinbothhospitalandremotesettings,despiteobstacles likeclassimbalanceandmisclassificationinunderrepresented conditions.

References

[1] \"Pulmo-TS2ONN:ANovelTripleScaleSelfOperational Neural Network for Pulmonary Disorder Detection Using RespiratorySounds,\"byA.Roy,U.Satija,andS.Karmakar, IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1–12, 2024, Art no. 6502812. [2] García-Ordás, M. T., Alaiz-Moretón, H., Benítez- Andrades, J. A., García-Rodríguez, I., & Benavides, C. (2024, February). Convolutional Neural Networks and Variational Autoencoders for Unbalancing Data in the Identification of Respiratory Pathologies. arXiv preprint arXiv:2402.02183. [3] Kavitha, M., Sreeja, S., Roopashri, G., Vidhya, K., & Muhil, P. (2024, February). RNN Framework-Based Automated Lung Disease Identification, Categorization, and Forecasting.pp.03015inE3SWebofConferences,vol.491. [4] Chen,Z.,Yeh,C.-H.,Wang,H.,&Liu,X.(2022, [5] August).IdentifyRespiratoryAbnormalitiesinLungSounds with a Fine-Tuned ResNet18 Network and STFT. arXiv preprint arXiv:2208.13943. [6] Mang, L. D., Garcia Galan, S., Martinez Munoz, D., Gonzalez Martinez, F. D., & Cortina, R. (2024, November). ClassificationofUnusualSoundsUsingVisionTransformers and Cochleograms. In arXiv preprint arXiv:2411.05955. [7] Satija,U.,andRoy,A.(2024).AnInnovativeMulti-Head Self-OrganizedOperationalNeuralNetworkStructureforthe Identification of Chronic Obstructive Pulmonary Disease ThroughLungSounds.pp.1–12inIEEE/ACMTransactions on Audio, Speech, and Language Processing, vol. 32 [8] Koppad, D., Kumar, P., Kantikar, N. A., K V, S., & Ramesh, S. (2024, April). Multi-Task Learning for ClassifyingLungSoundsandLungDiseases.ArXivpreprint arXiv:2404.03908. [9] Deeven,V.R.,Akshitha,N.,Sai,Y.P.,Kumar,V.N.,& Kaivalya, M. (2023, November). Deep Learning-Based PulmonarySoundAnalysisforEffectiveRespiratoryDisease Classification. pp. 1–7 in Proceedings of the Second InternationalConferenceonEmergingTrendsinEngineering (ICETE 2023). [10] Herasevich, S., Tekin, A., Pinevich, Y., Lipatov, K., Garcia-Mendez,J.P.,Lal,A.,Herasevich,S.,&Herasevich, [11] V. (2023, October). A Systematic Review of Machine Learning for Automated Classification of Abnormal Lung Sounds from Public Databases. Bioengineering, vol. 10, no. 10, pp. 1155 [12] Jiang, J., Wu, C., and Na, Y. E. (2024, September). A Convolutional Module for Spatial and Channel Reconstruction in a Lung Sound Classification Model. Southern Medical University Journal, vol. 44, no. 9, pp. 1720–1728. [13] Zhao,Z.,Gong,Z.,Niu,M.,Ma,J.,Wang,H.,Zhang,Z., & Li, Y. (2022, May). Classifying Respiratory Sounds Automatically with a Multi-Branch Temporal Convolutional Network. pp. 1–5 in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2022). [14] Sun, Z., and Wang, Z. (2024, April). Performance Assessment of Deep Learning-Based Lung Sound Classification with Variable Parameters. pp. 1–12 in EURASIP Journal on Advances in Signal Processing, vol. 2024, no. 51. [15] Rajadurai,P.,andS.Balasubramanian(2023).Real-time lung sound classification of pulmonary diseases using machine learning. pp. 122–130 in International Journal of Engineering and Technology Innovation, vol. 13, no. 3. [16] Islam, M. A., Bhattacharyya, P., Bandyopadhyaya, I., & Saha, G. (2018, April). Using multichannel lung sound signals, subjects with COPD, asthma, and normal can be categorized. pp. 290–294 in Proceedings of the International Conference on Communication and Signal Processing (ICCSP). [17] Rocha, B. M., Filos, D., Mendes, L., Vogiatzis, I., Perantoni, E., Kaimakamis, E.,... & Maglaveras, N. (2017, February). A Database of Respiratory Sounds for Automated Classification Development. 33–37 in Proceedings of the International Conference on Biomedical and Health Informatics (BHI).

Copyright

Copyright © 2025 Pooja M D, Nirmala S, Prarthana V Gunakimathi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.