Do you have COVID-19? An artificial intelligence-based screening tool for COVID-19 using acoustic parameters

Area of Science:

• Biomedical Engineering
• Computational Linguistics
• Infectious Disease Diagnostics

Background:

• COVID-19 diagnosis relies on methods like PCR and antigen tests.
• There is a need for accessible, non-invasive screening tools for early detection and public health monitoring.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based tool for screening COVID-19 using voice acoustic parameters.
• To assess the efficacy of machine learning models in identifying COVID-19 from voice data.

Main Methods:

• Voice samples (/a/ vowel phonation) were collected from 203 COVID-19 patients and 171 healthy individuals.
• Twenty-five acoustic parameters (e.g., fundamental frequency, harmonicity, airflow) were extracted.
• Feedforward Neural Network (FFNN) and other machine learning models were trained and validated using a leave-one-subject-out scheme.

Main Results:

• The FFNN model achieved high performance: 89.71% accuracy, 91.63% recall, and 90.62% F1-score.
• Logistic regression showed slightly higher precision (90.17%) than FFNN (89.63%).
• The FFNN model demonstrated robust performance for COVID-19 screening via voice analysis.

Conclusions:

• An AI-powered voice analysis tool can effectively screen for COVID-19.
• This technology offers a potential low-cost, non-invasive method for widespread respiratory health screening.
• The tool can identify potential abnormalities, prompting users to seek professional medical advice.