Lung parenchyma segmentation in CT scans as a preprocessing step for automatic nodule detection

“Computer Assisted Detection” systems (CADs) are software tools designed for assisting the physician both during usual clinical practice (as “2nd readers” of diagnostic analyses) and in screening programs (as 1st readers to sort patients, on the ground of disease evidence, before the physician examines the analyses). In both cases the diagnosis is made by the physician, who takes into account the information provided by the CAD system. The work presented in this paper was carried out in the framework of the INFN (Istituto Nazionale di Fisica Nucleare) funded MAGIC-5 (Medical Applications in a GRID Infrastructure Connection) project, whose aim is the development of software tools for medical applications, to be implemented in a GRID Virtual Organization. In particular we describe the initial step of a CAD system for malignant nodule detection in lung Computed-Tomography (CT) scans: the automated segmentation of the pulmonary parenchyma. A general overview of our segmentation algorithm is given; its overall structure is analyzed, and details about the innovative parts are illustrated.