Related Experiment Video
Updated: May 20, 2025
06:01
Semi-Automated Analysis of Peak Amplitude and Latency for Auditory Brainstem Response Waveforms Using R
Published on: December 9, 2022
2.4K
SympCoughNet: symptom assisted audio-based COVID-19 detection
Yuhao Lin1, Xiu Weng2, Bolun Zheng1
1School of Automation, Hangzhou Dianzi University, Hangzhou, China.
Frontiers in Digital Health
|March 27, 2025
Summary
This study introduces SympCoughNet, a deep learning model that uses cough sounds and clinical symptoms for COVID-19 detection. Integrating symptom data significantly improved accuracy over audio-only methods.
Area of Science:
- Medical Informatics
- Artificial Intelligence
- Respiratory Medicine
Background:
- COVID-19 diagnosis faces challenges with existing methods like nucleic acid tests, antigen tests, and CT imaging due to inefficiencies and limited accessibility.
- Current COVID-19 detection using acoustic health signals (cough, breathing sounds) often neglects valuable clinical symptom data, leading to suboptimal accuracy.
- There is a need for rapid, convenient, and accurate COVID-19 detection methods that leverage diverse data sources.
Purpose of the Study:
- To develop and evaluate SympCoughNet, a deep learning model for COVID-19 audio classification that integrates cough sounds with clinical symptom data.
- To enhance COVID-19 detection accuracy by incorporating symptom information through symptom-encoded channel weighting.
- To assess the impact of symptom integration on model performance through an ablation study.
Main Methods:
- Developed SympCoughNet, a deep learning network integrating cough sounds and clinical symptom data for COVID-19 audio classification.
- Employed symptom-encoded channel weighting to improve feature processing and attention to symptom information.
- Conducted an ablation study using a CNN-based architecture, treating symptoms as classification labels to evaluate symptom integration's impact.
Main Results:
- SympCoughNet achieved 89.30% accuracy, 94.74% AUROC, and 91.62% PR on the test set, outperforming traditional audio-only approaches.
- The integration of clinical symptom data significantly enhanced COVID-19 detection performance.
- The ablation study confirmed that the network effectively leverages cough audio to infer symptom-related information, even when symptoms are used as classification labels.
Conclusions:
- Integrating clinical symptom data with cough audio signals is a promising approach to improve COVID-19 detection accuracy.
- SympCoughNet demonstrates the potential of deep learning models to enhance diagnostic capabilities for infectious diseases.
- Further research should consider the impact of accurate symptom reporting on model predictions and explore broader applications of acoustic health signal analysis.
