PATIENT SPECIFIC MODELS IN LUNG CANCER SCREENING WITH CT

点击翻译按钮获取中文摘要

DESCRIPTION (Verbatim from Applicant's Abstract): The objective of this research is to develop computer-assisted methods to facilitate screening for the early detection of lung cancer using helical computed tomography (hCT). Proponents of existing screening trials argue that the highest enhance of surgical cure from lung cancer lies in the detection of micronodular neoplasms (of 1-3 mm in diameter). Multi-slice hCT is capable of imaging the entire thorax at high spatial resolution and has the potential to reliably detect pulmonary micronodules. However, these image sequences generate extremely large volume data sets, consisting of 300-600 axial images, that are impractical to review in current radiology practice. This proposal involves development and experimental testing of a method to automatically identify lung nodules from high resolution hCT (HR-hCT) image data acquired from multi-slice scanners. The technique involves a model-based segmentation approach in which information about the size, shape, location, density and other properties of both normal and pathological structures will be used to automate the discrimination of focal lung nodules from normal bronchovascular anatomy. A generic, a priori model of lung nodules and relevant anatomy will be developed to guide segmentation of baseline CT images. Patient-specific models will be derived from the anatomical information learned from baseline scans and used to analyze subsequent surveillance CT scans. The specific aims to accomplish this are: [1] To automatically distinguish lung nodules from normal pulmonary bronchovascular structures on baseline lung cancer screening HR-hCT exams. [2] To detect interval new nodules and re-localize previously detected nodules on post-baseline surveillance HR-hCT exams. [3] To measure the accuracy of automated nodule detection and re-localization on HR-hCT scans. [4] To compare radiologist accuracy and interpretation times of HR-hCT scans, both with and without assistance from the automated detection system, against pre-existing nodule detection methods.