Texture analysis improves lung-tissue segmentation on high-resolution computed tomography in COVID-19

Area of Science:

• Radiology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Accurate segmentation of lung parenchyma, ground-glass opacity (GGO), and intrapulmonary vessels in COVID-19 patients using high-resolution computed tomography (HRCT) is difficult.
• COVID-19 diagnosis and monitoring rely on precise identification of these lung abnormalities.

Purpose of the Study:

• To evaluate the effectiveness of texture-based feature extraction for classifying lung parenchyma, GGO, and intrapulmonary vessels in COVID-19 patients.
• To determine the optimal window size and most informative texture features for accurate classification.

Main Methods:

• A cross-sectional study analyzed 530 adults with confirmed COVID-19.
• 597 regions of interest (ROIs) representing parenchyma, GGO, and vessels were analyzed using first- and second-order texture features.
• Stepwise linear discriminant analysis was employed for feature selection, creating a three-class classifier.

Main Results:

• The 20x20-pixel window size yielded the highest classification performance with an overall accuracy of 88.6%.
• Five co-occurrence-based texture features were identified as most discriminative.
• Classification errors were predominantly observed at tissue boundaries with mixed voxels.

Conclusions:

• Texture-based feature extraction demonstrates high ROI-level accuracy (88.6%) for differentiating lung components in COVID-19.
• This approach can function as a valuable supplementary tool in the radiological interpretation of chest CT scans.