Pulmonary Artery-Vein Classification in CT Images Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Vascular Diseases

Background:

• Pulmonary vascular diseases affect arteries or veins via distinct mechanisms.
• Manual analysis of chest CT scans for vascular abnormalities is time-consuming and difficult to standardize.
• Accurate differentiation of arteries and veins in CT images is crucial for diagnosing pathological conditions.

Purpose of the Study:

• To develop and validate a fully automatic approach for classifying vessels as arteries or veins in chest CT images.
• To improve the efficiency and accuracy of diagnosing pulmonary vascular diseases.

Main Methods:

• A novel algorithm combining scale-space particles segmentation, a 3D convolutional neural network (CNN) for initial classification, and graph-cuts optimization for refinement.
• Comparison of various 2D and 3D CNN architectures and a random forests classifier.
• Training and evaluation on noncontrast chest CT scans, with validation on contrast-enhanced scans.

Main Results:

• The proposed method achieved an overall accuracy of 94%, outperforming other CNN architectures and random forests.
• The algorithm demonstrated generalization capabilities on contrast-enhanced CT scans from patients with chronic thromboembolic pulmonary hypertension.
• The approach significantly improves upon state-of-the-art methods for artery/vein classification in CT images.

Conclusions:

• The developed fully automatic method effectively classifies arteries and veins in chest CT images.
• This 3D CNN-based approach offers a more accurate and efficient alternative to manual analysis for diagnosing pulmonary vascular diseases.
• The method shows promise for future clinical applications in aiding disease diagnosis and management.