Clinical Report: AI Algorithm for COVID-19 Detection via Chest CT Validated by ICOVAI

Overview

The ICOVAI consortium developed and externally validated an AI model for COVID-19 detection and lung involvement quantification on chest CT scans using a large, multicenter European dataset. The model demonstrated robust performance in automated CO-RADS classification and lung opacity segmentation, supporting its potential clinical utility.

Background

During the COVID-19 pandemic, chest CT imaging has been critical for evaluating lung involvement and disease severity. Manual assessment of lung abnormalities is time-consuming, prompting development of AI tools to automate segmentation and CO-RADS classification. CO-RADS standardizes reporting of COVID-19 probability on CT, but interobserver variability remains a challenge. External validation of AI models is essential to ensure generalizability and clinical applicability.

Data Highlights

Key Findings

• The AI model was trained on 1286 CT scans from 1266 patients across 10 institutions, including balanced COVID-19 positive and negative cases.
• External validation was performed on 375 CT scans from 5 European hospitals, with RT-PCR as the reference standard.
• Multiple expert radiologists independently annotated CO-RADS scores and lung segmentations to ensure high-quality ground truth data.
• Discordant CO-RADS cases were excluded to improve training label reliability.
• The AI model performed automated lung volume and opacity segmentation and CO-RADS classification with high accuracy.
• Preprocessing included voxel resampling and intensity normalization to standardize input data for the AI model.

Clinical Implications

The validated AI model can assist radiologists by providing rapid, standardized CO-RADS classification and quantitative lung involvement metrics at the time of CT interpretation. This may enhance diagnostic consistency, reduce workload, and improve clinical workflow efficiency, especially in high-volume or resource-limited settings. External validation across multiple centers supports the model's generalizability for broader clinical adoption.

Conclusion

The ICOVAI AI algorithm demonstrates reliable performance for automated COVID-19 detection and lung involvement quantification on chest CT, validated on a large, diverse external dataset. This supports its potential integration into clinical practice to aid radiological assessment during the pandemic.

References

1. Roberts et al 2021 -- Systematic review on AI in COVID-19 imaging
2. CO-RADS introduction and validation studies