Computer Aided Diagnostic Lung Cancer Detection Classification Computed Tomograph

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Early detection and diagnosis of lung cancer are very important because they may lead to an improved prognosis for lung cancer patients. Our goal in this project is to develop and evaluate an integrated interactive computer-aided diagnostic (CAD) scheme for early detection and diagnosis of lung cancer in multi-slice computed tomography (CT), which will incorporate a detection scheme and a diagnosis scheme into a clinically oriented system. The integration of the two CAD schemes will provide radiologists with a highly practical and convenient diagnostic environment, and it will also provide technically improved reliability and compactness compared with the use of two separate CAD schemes. The interaction between radiologists and a CAD scheme enables radiologists to correct the computer's detection errors, such as to identify cancers missed by the computer, and to ignore most of the non-nodules incorrectly detected by the computer. As a result, it is expected that radiologists would become more accurate and productive in both detection and diagnosis of lung cancers by use of the integrated CAD scheme. Many innovative techniques and schemes are to be developed in this project, including (1) an accurate segmentation technique based on spiral scanning and dynamic programming; (2) an automated rule-based classifier using multiple composite features with minimized overtraining effects; (3) a nodule detection scheme using the nodule segmentation technique, the automated rule-based classifier, and a selective nodule enhancement filter; (4) a three-category nodule diagnosis scheme for distinguishing cancers from benign nodules and nonnodules; and (5) integration of the detection and diagnosis schemes, with the capability of interaction between the CAD scheme and its users. It is anticipated that the computerized detection scheme will detect 90% of nodules, with less than 4 non-nodules per CT scan, and that the diagnosis scheme will achieve an Az value (area under the receiver operating characteristic curve) of 0.90 or higher for distinction between cancers and non-cancers. Finally, an observer performance study will be conducted with 20 participating radiologists to examine the clinical usefulness of the integrated interactive CAD scheme. It is expected from the observer study that radiologists would significantly improve their detection and diagnosis accuracy for lung cancer with the aid of the interactive CAD scheme.