Automatic detection of pulmonary arteries and assessment of bronchial dilatation in HRCT images of the lungs

Abstract

Bronchial dilatation is one of the most important direct signs for the diagnosis of bronchiectasis in high-resolution CT images of the lung. The assessment of the dilatation is done by comparing the size of the bronchus and accompanying artery. Previous work has shown that the success of an automatic bronchial dilatation detection method is limited by high measurement error rate of small bronchi and arteries. This paper presents a new method for automatic detection of accompanying arteries and assessment of bronchial dilatation. A knowledge-guided template matching is used to approximately locate the accompanying artery of a bronchus. A seeded region growing, with leaking prevention and correction, is used to precisely segment the artery. Bronchus-artery lumen area ratio (LAR) and their shortest diameter ratio (SDR) are used to compare the sizes of a bronchus and the accompanying artery. Machine learning is used to determine the suitable severity thresholds for different sizes of bronchi. The method was evaluated using 324 images from 64 patient studies. The results were compared with manual identification and classification, which were verified by an experienced radiologist. The method achieved 90% and 82% accuracies for artery detection and dilatation assessment, respectively