Computing infection distributions and longitudinal evolution patterns in lung CT images.

Area of Science:

• Medical Imaging Analysis
• Radiology
• Computational Pathology

Background:

• Understanding spatial and temporal lung infection patterns in coronavirus disease 2019 (COVID-19) is crucial for disease comprehension and management.
• Existing methods lack automated pipelines for statistically analyzing infection distributions and their evolution over time.

Purpose of the Study:

• To develop and validate an automated pipeline for segmenting and registering lung infection regions in CT images.
• To statistically analyze and visualize spatial and temporal infection patterns in COVID-19 patients.
• To compare infection patterns between COVID-19 and community-acquired pneumonia (CAP), and between different severity levels of COVID-19.

Main Methods:

• A VB-Net model was developed for automated segmentation of infection regions in CT scans.
• Deformable registration was employed to align segmented infections onto a common template.
• Voxel-level analysis was used to compute spatial distributions and track changes during disease progression, enabling comparisons between patient groups.

Main Results:

• The VB-Net achieved high segmentation accuracy (Dice 91.6% ± 10.0%), comparable to inter-radiologist agreement.
• Infection probability was highest in peripheral subpleural regions; COVID-19 GGO lesions were more widespread than consolidations.
• Distinct patterns were observed between COVID-19 and CAP, severe and critical COVID-19, and four progression patterns were identified in critical cases.

Conclusions:

• The automated pipeline effectively visualizes and quantifies spatial infection patterns and disease course changes.
• The findings highlight significant differences in infection distribution and progression between COVID-19 and CAP, and varying severity levels.
• The study provides a valuable tool for understanding lung infection dynamics and offers insights into COVID-19's multifaceted presentation.