Quantitative Assessment of COVID-19 Lung Disease Severity: A Segmentation-based Approach

Area of Science:

• Radiology
• Medical Imaging Analysis
• Computational Pathology

Background:

• Accurate lung segmentation is crucial for quantitative analysis in respiratory diseases.
• Existing image analysis models require robust evaluation metrics for clinical utility.

Purpose of the Study:

• To introduce mean Hounsfield units (HU) as a quantitative metric for respiratory disease severity.
• To compare the performance of a novel 3D global context attention network (GC-Net) against the V-Net segmentation algorithm for lung CT scans.

Main Methods:

• Utilized thoracic High-Resolution Computed Tomography (HRCT) scans from patients with COPD and COVID-19.
• Implemented a novel 3D global context attention network (GC-Net) for lung segmentation.
• Employed a biomimetic data augmentation strategy and mean HU as a quantitative severity metric.

Main Results:

• Mean HU effectively quantified disease severity, enabling detailed comparison between GC-Net and V-Net segmentation models.
• The study analyzed the strengths and weaknesses of implemented models within the context of respiratory disease lung segmentation.

Conclusions:

• Mean Hounsfield units provide an objective measure for respiratory disease severity.
• This metric is valuable for comparing the performance of CT scan analysis algorithms in clinical settings.