Deep learning models-based CT-scan image classification for automated screening of COVID-19

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• COVID-19, caused by SARS-CoV-2, is a highly transmissible respiratory illness impacting millions globally.
• Accurate and timely diagnosis is crucial for controlling the spread of COVID-19.
• Chest CT scans are valuable tools for diagnosing respiratory conditions, including COVID-19.

Purpose of the Study:

• To develop a robust framework for automated COVID-19 screening using chest CT images.
• To leverage deep learning models for efficient and accurate detection of COVID-19.
• To reduce the burden on clinicians and improve patient outcomes through automated analysis.

Main Methods:

• Utilized a public dataset of 1252 COVID-19 and 1230 non-COVID chest CT images.
• Employed pre-trained deep learning models (MobileNetV2, DarkNet19) and a novel lightweight DLM.
• Implemented a repeated ten-fold cross-validation strategy for model training and testing.

Main Results:

• The transfer-learned DarkNet19 model achieved the highest classification accuracy of 98.91%.
• The proposed framework demonstrated high effectiveness in automated COVID-19 screening.
• Simulation results validated the performance on a publicly available dataset.

Conclusions:

• Automated COVID-19 detection using deep learning on chest CT scans is highly accurate.
• The developed framework shows promise for clinical application in early COVID-19 diagnosis.
• Further testing with larger datasets is recommended to enhance the framework's generalizability.