Cough Audio Analysis for COVID-19 Diagnosis

Area of Science:

• Medical diagnostics
• Artificial intelligence in healthcare
• Infectious disease research

Background:

• COVID-19 pandemic caused significant global health challenges.
• Reverse-transcription polymer chain reaction (RT-PCR) is a standard COVID-19 diagnostic but is time-consuming and lacks scalability.
• Need for rapid, scalable diagnostic methods for COVID-19 is critical.

Purpose of the Study:

• To develop a preliminary, scalable, and less time-consuming method for COVID-19 prognosis.
• To competitively analyze the performance of four distinct machine learning models for COVID-19 diagnosis.
• To identify the most effective machine learning model for distinguishing between positive and negative COVID-19 cough sounds.

Main Methods:

• Comparative analysis of four machine learning models: Multilayer Perceptron (MLP), Convolutional Neural Networks (CNN), Recurrent Neural Networks with Long Short-Term Memory (LSTM-RNN), and VGG-19 with Support Vector Machines (SVM).
• Utilized a robust feature embedding technique to differentiate between COVID-19 positive and negative cough sounds.
• Evaluated model performance based on accuracy, specificity, and scalability.

Main Results:

• Multilayer Perceptron demonstrated superior performance compared to CNN, LSTM-RNN, and VGG-19 with SVM.
• Multilayer Perceptron achieved a high specificity of 94.5% and an accuracy of 96.8%.
• The chosen feature embedding technique effectively enabled the MLP model to distinguish between COVID-19 positive and negative coughs.

Conclusions:

• Machine learning, specifically Multilayer Perceptron, offers a promising, rapid, and scalable approach for COVID-19 diagnosis.
• The developed method using MLP and feature embedding provides a viable alternative to traditional RT-PCR testing.
• Further research can explore the clinical application of this AI-driven diagnostic tool for early COVID-19 detection.