The broad, long term objective of the proposed research is to develop a fully automated, computerized system that will assist radiologists in the detection and quantitative assessment of pulmonary nodules in helical computed tomography (CT) images of the thorax. This system will potentially improve the prognosis of patients with lung cancer by contributing to earlier diagnosis. It is widely recognized that helical CT is the most sensitive imaging modality for the valuation of lung nodules. The large amount of image data acquired during a CT scan, however, makes nodule detection by human observers a difficult task. Moreover, distinguishing between nodules and normal anatomy such as pulmonary vessels typically requires visual comparison among multiple CT sections, each of which contains information that must be evaluated by a radiologist and assimilated into the larger context of the volumetric data acquired during the scan. This evaluation requires the radiologist to mentally construct a three-dimensional representation of patient anatomy based on over 50 section images acquired during a CT examination. This task, while cumbersome for radiologists, may be efficiently handled by a computerized method. The proposed research project will investigate the two-dimensional and three-dimensional structure of lung nodules in helical CT images to fully exploit the volumetric image data acquired during a CT examination. Gray-level threshold-based techniques will be used to extract three-dimensional structures from CT image data. Quantitative geometric and gray-level information computed for nodule candidates will be used as input to automated classifiers to distinguish between structures that correspond to nodules and structures that correspond to normal anatomy. This quantitative information will also allow for an evaluation of detection performance based on radiologic appearance of nodules.
The specific aims of the proposed research are: (1) to collect databases of normal and abnormal helical thoracic CT scans, (2) to develop an automated method to detect and quantitatively assess pulmonary nodules in these CT scans, (3) to investigate differences in the appearance of nodules imaged in low-dose helical thoracic CT scans obtained from a lung cancer screening program as opposed to standard helical CT and the effect of these differences on the detection scheme, and (4) to evaluate the performance of the computerized detection scheme and its effect on the performance of radiologists in the task of identifying pulmonary nodules.
