Lung cancer is the most deadly cancer in both men and women. This can be largely attributed to the fact that lung cancer is usually detected in a late stage. If the disease is detected in an early stage, the survival rate is much better. Therefore, early detection of lung cancer, in which it is still treatable, is of major importance to reduce lung cancer mortality.udEarly stage lung cancer manifests itself as pulmonary nodules, which are described as round opacities, well or poorly defined, measuring up to 3 cm in diameter. Thin-slice helical chest CT scans have a sub-millimeter resolution at which small pulmonary nodules can be detected. Computer-aided detection of lung nodules has the potential to increase reader sensitivity for the detection of pulmonary nodules and may reduce reading time. Furthermore, automated characterization of pulmonary nodules may assist the radiologist in assessing the likelihood of malignancy of lung nodules.udIn this thesis, novel detection and characterization systems for pulmonary nodules are described. We proposed a novel subsolid CAD system which aims to detect subsolid nodules, a system to detect and quantify micronodules, and a system to automatically detect interval change between consecutive CT scans. All three systems were evaluated on large datasets and showed promising performance. In addition, we performed a comparative study with three CAD algorithms on the largest publicly available reference database for pulmonary nodules. Next, we described a method which automatically classifies pulmonary nodules into solid, part-solid, or non-solid nodules. This is crucial for selecting the appropriate workup for pulmonary nodules. Finally, we discussed how the developed methods can be efficiently integrated into clinical practice.
展开▼
