A rapid screening classifier for diagnosing COVID-19.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Infectious Diseases

Background:

• The COVID-19 pandemic necessitated rapid diagnostic tools.
• Chest X-ray (CXR) and clinical data offer potential for screening.
• Distinguishing COVID-19 from influenza pneumonia is clinically important.

Purpose of the Study:

• To develop and evaluate a classifier combining CXR and clinical features for COVID-19 diagnosis.
• To compare the classifier's performance against clinical features alone, CXR alone, and experienced physicians.
• To assess the classifier's utility in differentiating COVID-19 from influenza A/B pneumonia.

Main Methods:

• A deep neural network (DNN) was trained on fused features from CXR and clinical data of 512 COVID-19 patients and 106 influenza patients.
• Performance was evaluated using Area Under the Receiver Operating Curve (AUC), sensitivity, and specificity.
• A reader study compared the AI system's diagnostic performance against three experienced pulmonary physicians.

Main Results:

• The combined classifier achieved an AUC of 0.952 (91.5% sensitivity, 81.2% specificity) for differentiating COVID-19 from influenza.
• COVID-19 patients showed distinct clinical patterns: less fever, more diarrhea, and higher hypercoagulability.
• The AI system outperformed experienced physicians in diagnostic accuracy and turnaround time.

Conclusions:

• The developed classifier is an efficient, economical, and radiation-safe tool for rapid COVID-19 screening.
• This approach effectively distinguishes COVID-19 from influenza A/B pneumonia.
• The AI-powered classifier shows significant promise as a frontline diagnostic tool during pandemics.