Abstract
In the last two decades, several methods for airway segmentation from chest CT images have been proposed. The following natural step is the development of a tool to accurately assess the morphology of the bronchial system in all its aspects to help physicians better diagnosis and prognosis complex pulmonary diseases such as COPD, chronic bronchitis and bronchiectasis. Traditional methods for the assessment of airway morphology usually focus on lumen and wall thickness and are often limited due to resolution and artifacts of the CT image. Airway wall cartilage is an important characteristic related to airway integrity that has shown to be deteriorated during the airway disease process. In this paper, we propose the development of a Model-Based GAN Regressor (MBGR) that, thanks to a model-based GAN generator, generate synthetic airway samples with the morphological components necessary to resemble the appearance of real airways on CT at will and that simultaneously measures lumen, wall thickness, and amount of cartilage on pulmonary CT images. The method is evaluated by first computing the relative error on generated images to show that simulating the cartilage helps improve the morphological quantification of the airway structure. We then propose a cartilage index that summarizes the degree of cartilage of bronchial trees structures and perform an indirect validation with subjects with COPD. As shown by the results, the proposed approach paves the way for the use of CNNs to precisely and accurately measure small lung airways morphology, with the final goal to improve the diagnosis and prognosis of pulmonary diseases.