Volume 28, Issue 2 (9-2026)
yafte 2026, 28(2): 93-109 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Ghaffari H, Dalvand S. A COVID-19 Diagnostic Approach Based on Cough Sound Analysis Using Deep Learning with Convolutional Neural Networks and Long Short-Term Memory Networks. yafte 2026; 28 (2) :93-109
URL: http://yafte.lums.ac.ir/article-1-3881-en.html
URL: http://yafte.lums.ac.ir/article-1-3881-en.html
Department of Computer Engineering, Fe.C., Islamic Azad University, Ferdows, Iran
Abstract: (10 Views)
Background: Conventional COVID-19 diagnostic assays, such as reverse transcription polymerase chain reaction (RT-PCR), may be time-consuming, costly, and susceptible to environmental and operational factors. These limitations are particularly challenging in resource-constrained settings. Artificial intelligence-based analysis of cough sounds has emerged as a promising noninvasive approach for COVID-19 screening.
Materials and Methods: This study proposes a novel COVID-19 detection method based on cough sound analysis. First, cough signals are converted into spectrogram images. Next, spatial features are extracted using an enhanced Convolutional Neural Network (CNN), the hyperparameters of which are optimized using the Coati Optimization Algorithm. The Arithmetic Optimization Algorithm is then employed to select the most informative features. Finally, a Long Short-Term Memory (LSTM) network classifies the samples into COVID-19-infected and healthy groups.
Results: On the CHRSD dataset, the proposed method achieved an accuracy of 95.72%, a sensitivity of 93.41%, and a precision of 91.75%. On the Cambridge dataset, it achieved an accuracy of 98.26%, a sensitivity of 97.73%, and a precision of 97.92%. On the COUGHVID dataset, the method obtained an accuracy of 98.83%, a sensitivity of 97.29%, and a precision of 97.68%.
Conclusion: The proposed method outperformed conventional CNN, Deep Neural Network, and LSTM models in COVID-19 detection based on cough sounds. Its noninvasive nature and computational efficiency indicate its potential for deployment in smartphone-based screening applications. However, the method should be considered a supportive screening tool and requires further clinical validation before use as a replacement for standard diagnostic tests, such as RT-PCR.
Materials and Methods: This study proposes a novel COVID-19 detection method based on cough sound analysis. First, cough signals are converted into spectrogram images. Next, spatial features are extracted using an enhanced Convolutional Neural Network (CNN), the hyperparameters of which are optimized using the Coati Optimization Algorithm. The Arithmetic Optimization Algorithm is then employed to select the most informative features. Finally, a Long Short-Term Memory (LSTM) network classifies the samples into COVID-19-infected and healthy groups.
Results: On the CHRSD dataset, the proposed method achieved an accuracy of 95.72%, a sensitivity of 93.41%, and a precision of 91.75%. On the Cambridge dataset, it achieved an accuracy of 98.26%, a sensitivity of 97.73%, and a precision of 97.92%. On the COUGHVID dataset, the method obtained an accuracy of 98.83%, a sensitivity of 97.29%, and a precision of 97.68%.
Conclusion: The proposed method outperformed conventional CNN, Deep Neural Network, and LSTM models in COVID-19 detection based on cough sounds. Its noninvasive nature and computational efficiency indicate its potential for deployment in smartphone-based screening applications. However, the method should be considered a supportive screening tool and requires further clinical validation before use as a replacement for standard diagnostic tests, such as RT-PCR.
Keywords: COVID-19, Cough sound analysis, Deep learning, Lung infection, Signal processing, Swarm intelligence
Type of Study: Original Research |
Subject:
عفونی
Received: 2025/11/29 | Accepted: 2026/06/14 | Published: 2026/09/1
Received: 2025/11/29 | Accepted: 2026/06/14 | Published: 2026/09/1
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
