Clinical Scorecard: Assessing the Clinical Applicability of an AI Algorithm for COVID-19 Detection via Chest CT: Findings from the International Consortium for COVID-19 Imaging AI (ICOVAI)

At a Glance

Key Highlights

• Automated AI segmentation correlates with disease severity by quantifying affected lung tissue on chest CT.
• CO-RADS classification standardizes COVID-19 imaging reporting with five categories indicating disease probability.
• External validation on independent multicenter datasets is critical to ensure AI model generalizability and clinical applicability.

Guideline-Based Recommendations

Diagnosis

• Use RT-PCR testing within 7 days of imaging as reference standard for COVID-19 diagnosis.
• Apply CO-RADS classification to standardize chest CT reporting for COVID-19 suspicion.

Management

• Incorporate AI-based automated segmentation and CO-RADS classification to assist radiologists in interpreting chest CT scans.
• Use AI outputs to potentially improve workflow efficiency and diagnostic consistency, especially for less experienced readers.

Monitoring & Follow-up

• Perform consensus reading by experienced radiologists in cases of discordant CO-RADS scores or uncertainty.
• Review AI segmentations by radiologists to ensure accuracy before clinical use.

Risks

• Be aware of potential methodological flaws and biases in AI models without external validation.
• Exclude poor quality CT scans (e.g., motion artifacts, insufficient inspiration) to maintain diagnostic accuracy.

Patient & Prescribing Data

Adult patients undergoing chest CT for suspected COVID-19 pneumonia or triage

AI model trained and validated on a large, diverse multicenter dataset with balanced COVID-19 positive and negative cases to ensure robust performance.

Clinical Best Practices

• Use high-quality, diverse, multicenter datasets with expert annotations for AI model development.
• Perform external validation on independent datasets before clinical implementation of AI tools.
• Combine AI outputs with expert radiologist review to optimize diagnostic accuracy and clinical workflow.
• Exclude scans with technical limitations or artifacts to maintain model reliability.

References

• Roberts et al systematic review on AI in COVID-19 imaging
• CO-RADS: COVID-19 Reporting and Data System