Particle Swarm Optimization-Based Extreme Learning Machine for COVID-19 Detection.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Respiratory Medicine

Background:

• COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Machine learning (ML) shows promise in analyzing respiratory sounds for disease detection.
• Existing ML models for COVID-19 detection often use limited voice datasets, excluding speech and vowels.

Purpose of the Study:

• To propose and evaluate a novel COVID-19 detection system utilizing ML.
• To assess the efficacy of the particle swarm optimization-extreme learning machine (PSO-ELM) algorithm for COVID-19 detection using diverse respiratory sounds.
• To expand COVID-19 detection capabilities beyond cough and breath sounds to include speech and vowel phonations.

Main Methods:

• Feature extraction using Mel Frequency Cepstral Coefficients (MFCCs).
• Classification using the Particle Swarm Optimization-Extreme Learning Machine (PSO-ELM) algorithm.
• Utilized the Corona Hack Respiratory Sound Dataset (CHRSD) encompassing thirteen distinct respiratory sound scenarios (breaths, coughs, speech, vowels).

Main Results:

• The PSO-ELM model achieved high detection accuracies across various scenarios.
• Highest accuracies included 96.43% for 'cough heavy' and 'count normal', and 96.15% for 'count fast', 'vowel a', and 'vowel e'.
• The system demonstrated robust performance, with accuracies ranging from 82.89% ('all vowels') to 96.43%.

Conclusions:

• The PSO-ELM algorithm is an efficient and accurate technique for COVID-19 detection using respiratory voice data.
• Incorporating a wider range of respiratory sounds, including speech and vowels, enhances the potential of ML-based diagnostic systems.
• This approach offers a promising non-invasive method for augmenting COVID-19 diagnosis.