COVID-19 activity screening by a smart-data-driven multi-band voice analysis.

Area of Science:

• Biomedical Engineering
• Data Science
• Respiratory Medicine

Background:

• The COVID-19 pandemic caused millions of deaths globally.
• Accurate, non-invasive, and low-cost diagnostic methods for COVID-19 are crucial for disease control.
• Existing diagnostic methods may have limitations in speed and accessibility.

Purpose of the Study:

• To develop and evaluate a novel method for COVID-19 diagnosis using non-linear analysis of cough, breathing, and speech signals.
• To assess the effectiveness of an XGBoost classifier in discriminating COVID-19 stages based on extracted signal patterns.
• To compare the proposed method's performance against existing state-of-the-art techniques.

Main Methods:

• Applied three signal analysis techniques (broadband, sub-bands, broadband & sub-bands) to cough, breathing, and speech signals from the Coswara dataset.
• Extracted non-linear patterns including Energy, Entropies, Correlation Dimension, Detrended Fluctuation Analysis, Lyapunov Exponent, and Fractal Dimensions.
• Utilized an XGBoost classifier to differentiate between COVID-19 positive cases and healthy controls across different disease stages.

Main Results:

• Achieved classification accuracies ranging from 83.33% to 98.46%.
• Demonstrated superior performance compared to state-of-the-art methods in certain comparisons.
• Attained a high accuracy of 98.46% when distinguishing between healthy controls and all COVID-19 stages.

Conclusions:

• The proposed signal analysis method shows significant promise for COVID-19 diagnosis screening.
• The technique offers a potentially accurate, non-invasive, and low-cost approach to assist in disease detection.
• Further validation may establish this method as a valuable tool for public health surveillance and early diagnosis.