Pleural nodule identification in low-dose and thin-slice lung computed tomography

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Pleural nodules are significant findings in lung computed tomography (CT).
• Accurate identification of these nodules is crucial for diagnosis and treatment planning.
• Automated systems can aid radiologists in detecting subtle findings.

Purpose of the Study:

• To develop and validate a fully automated system for pleural nodule identification in low-dose, thin-slice lung CT.
• To assess the system's performance using morphological and textural features analyzed by a neural classifier.
• To investigate the impact of ground truth agreement levels on system sensitivity.

Main Methods:

• A directional-gradient concentration method combined with morphological opening was used to detect nodule candidates on the pleura.
• Nodule candidates were characterized by 12 morphological and textural features.
• A rule-based filter and a neural classifier, evaluated using k-fold cross-validation, analyzed these features.

Main Results:

• The automated system was developed and validated on 42 annotated CT scans.
• Sensitivity varied significantly with ground truth agreement: 44% (1 radiologist) vs. 72% (2 radiologists).
• False positives were approximately six per scan at the lower agreement level.

Conclusions:

• A fully automated system for pleural nodule detection in lung CT has been successfully developed.
• System performance is sensitive to the level of consensus in the ground truth annotation.
• Further refinement may be needed to optimize performance across varying annotation standards.