CT-based severity assessment for COVID-19 using weakly supervised non-local CNN.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate patient criticality evaluation is crucial for effective COVID-19 treatment.
• Artificial Intelligence (AI) models can automate risk-stratification using clinical data.
• Chest CT findings like ground-glass opacities and consolidations correlate with disease severity.

Purpose of the Study:

• To develop a novel attention framework for estimating COVID-19 severity using weakly annotated chest CT scans.
• To create an AI model that provides a regression score for patient risk stratification.

Main Methods:

• A non-locality attention approach correlating features across 3D scan parts and scales.
• An explicit guidance mechanism using limited infection labeling for attention refinement.
• Cross-channel attention and global contextual awareness infusion into voxel features.

Main Results:

• The proposed attention framework achieved an R-squared score of 0.84.
• The model demonstrated a mean absolute difference of 0.133 on the MosMed dataset.
• The architecture effectively localized infection regions and refined features.

Conclusions:

• The novel attention framework shows significant potential for augmenting COVID-19 severity assessment.
• AI-driven analysis of chest CT scans can improve patient risk stratification.
• The model precisely identifies and quantifies infection indicators for better prognostic studies.