Mini-COVIDNet

Mini-COVIDNet classifies lung ultrasound images to detect and distinguish COVID-19, pneumonia, and healthy lung conditions for point-of-care diagnostic use.

Key Features:

• Lightweight Architecture: Has 4.39 times fewer parameters than its closest competitor and requires 51.29 MB of memory, enabling deployment on embedded platforms and mobile devices.
• High Accuracy: Achieves an accuracy rate of 83.2% in distinguishing COVID-19, pneumonia, and healthy lung conditions from lung ultrasound images.
• Rapid Training: Requires 24 minutes to train, reducing time to iteration and deployment.

Scientific Applications:

• Point-of-care lung ultrasound diagnosis: Leverages lung ultrasound imaging, which requires minimal personal protective equipment and straightforward disinfection, for rapid screening and triage of COVID-19, pneumonia, and healthy lungs.

Methodology:

Benchmarked against lightweight and state-of-the-art heavy neural network models to compare accuracy, computational resource usage, and latency.