Automated lung nodule classification following automated nodule detection on CT: A serial approach

• DOI: 10.1118/1.1573210
• Publication Date: 2003-06-13T13:59:06Z
• Authors: Samuel G. Armato; Michael B. Altman; Joel Wilkie; Shusuke Sone; Feng Li; Kunio Doi; Arunabha S. Roy
• Keywords: Nodule (geology); Solitary pulmonary nodule

We have evaluated the performance of an automated classifier applied to task differentiating malignant and benign lung nodules in low-dose helical computed tomography (CT) scans acquired as part a cancer screening program. The classified this manner were initially identified by our nodule detection method, so that output was used input classification. This study begins narrow distinction between "detection task" "classification task." Automated is based on two- three-dimensional analyses CT image data. Gray-level-thresholding techniques are identify initial candidates, for which morphological gray-level features computed. A rule-based approach reduce number candidates correspond non-nodules, remaining merged through linear discriminant analysis obtain final results. classification merges algorithm actual another distinguish nodules. method computerized results obtained from database 393 thoracic containing 470 confirmed (69 401 nodules). Receiver operating characteristic (ROC) evaluate ability differentiate lesions. area under ROC curve attained value 0.79 during leave-one-out evaluation.