Optimized DEC: An effective cough detection framework using optimal weighted Features-aided deep Ensemble classifier...

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Signal Processing

Background:

• COVID-19 diagnosis is challenging due to symptom variability and lengthy processing times of existing methods.
• Cough audio signals present a promising, non-invasive biomarker for early COVID-19 detection.
• Developing efficient and accurate screening tools is crucial for pandemic management.

Purpose of the Study:

• To propose a novel ensemble-based deep learning model for accurate COVID-19 detection using cough audio signals.
• To enhance early screening and diagnosis by overcoming limitations of current methods.
• To investigate the efficacy of combining signal decomposition, feature extraction, and optimized deep learning classifiers.

Main Methods:

• Cough audio signals were decomposed using Empirical Mean Curve Decomposition (EMCD).
• Extracted features included Mel Frequency Cepstral Coefficients (MFCC), spectral, and statistical features.
• A Modified Cat and Mouse Based Optimizer (MCMBO) was used to optimize feature weighting and deep ensemble classifier (ODEC) parameters, which integrated Radial Basis Function (RBF), Long-Short Term Memory (LSTM), and Deep Neural Network (DNN) classifiers.

Main Results:

• The proposed model achieved an accuracy of 96% and a precision of 92%.
• Feature fusion and optimization using MCMBO significantly improved detection performance.
• The Optimized Deep Ensemble Classifier (ODEC) demonstrated robust diagnostic capabilities.

Conclusions:

• The developed deep learning model effectively detects COVID-19 from cough audio signals.
• This approach offers a rapid and accurate method for early COVID-19 screening.
• The findings support the use of AI-powered audio analysis in clinical practice for infectious disease diagnosis.