A deep learning algorithm using CT images to screen for Corona virus disease (COVID-19).

Area of Science:

• Radiology
• Artificial Intelligence
• Infectious Disease

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Pathogen testing for COVID-19 has limitations, including significant false negativity.
• Early detection is crucial for effective quarantine and treatment to control disease spread.

Purpose of the Study:

• To investigate the potential of artificial intelligence (AI) in analyzing CT images for COVID-19 diagnosis.
• To develop and validate an AI algorithm capable of identifying specific radiological features of COVID-19.
• To assess if AI can provide a diagnosis preceding pathogenic tests, thereby saving critical time.

Main Methods:

• A dataset of 1065 CT images from confirmed COVID-19 cases and viral pneumonia cases was collected.
• An inception transfer-learning model was modified to create the AI algorithm.
• The algorithm underwent both internal and external validation using distinct datasets.

Main Results:

• Internal validation demonstrated an overall accuracy of 89.5% (sensitivity 0.87, specificity 0.88).
• External validation yielded an overall accuracy of 79.3% (sensitivity 0.67, specificity 0.83).
• The AI correctly identified 85.2% of COVID-19 cases with initially negative nucleic acid tests.

Conclusions:

• AI can effectively extract radiological features from CT images for COVID-19 diagnosis.
• The developed AI model shows promise as a screening tool, particularly during influenza seasons.
• This AI approach can aid in differentiating COVID-19 from other viral pneumonias with similar radiographic characteristics.