Usefulness of computerized scheme for differentiating benign from malignant lung nodules on high-resolution CT Feng Li a, * , Qiang Li a , Masahito Aoyama b , Junji Shiraishi a , Hiroyuki Abe a , Kenji Suzuki a , Roger Engelmann a , Shusuke Sone c , Heber MacMahon a , Kunio Doi a a Kurt Rossmann Laboratories for Radiologic Image Research, Department of Radiology, The University of Chicago, MC-2026, 5841 South Maryland Avenue, Chicago, IL 60637, USA b Department of Intelligent Systems, Faculty of Information Sciences, Hiroshima City University, Hiroshima 731-3194, Japan c Azumi General Hospital, Ikeda, Nagano 399-8695, Japan Abstract. A computer-aided diagnosis (CAD) scheme for determination of the likelihood of malignancy of 244 nodules on high-resolution CT (HRCT) was developed. The performance (A z ) for 16 radiologists was improved from 0.785 to 0.853 ( P = 0.02) with the aid of the CAD scheme by use of 56 nodules, including 28 cancerous and 28 benign nodules which were matched in size and pattern to the cancers. Our purpose in this study was to investigate further whether a CAD scheme can assist radiologists in distinguishing benign from malignant nodules in different groups. The results indicated that A z values for radiologists without and with the CAD scheme were improved from 0.770 to 0.855 for general radiologists ( P = 0.01) and from 0.805 to 0.850 for chest radiologists ( P = 0.12); from 0.717 to 0.821 for nodules at 6–10 mm ( P = 0.04) and from 0.837 to 0.901 for nodules at 11–20 mm ( P = 0.04); and from 0.812 to 0.892 for nodules with pure ground-glass opacity (GGO) ( P = 0.149), from 0.819 to 0.863 for nodules with mixed GGO ( P = 0.196), and from 0.784 to 0.844 for solid nodules ( P = 0.334). CAD has the potential to improve the diagnostic accuracy in distinguishing benign nodules from malignant ones in different groups on HRCT. D 2004 Published by Elsevier B.V. Keywords: Lung nodule; Computer-aided diagnosis (CAD); Receiver operating characteristic (ROC) curve 1. Introduction Low-dose spiral CT has been applied for cancer screening and led to early detection of lung cancer in some countries [1–3]. However, simultaneous or additional diagnostic high-resolution CT (HRCT) was needed for distinction between early cancers, most of 0531-5131/ D 2004 Published by Elsevier B.V. doi:10.1016/j.ics.2004.03.083 * Corresponding author. Tel.: +1-773-834-5093; fax: +1-773-702-0371. E-mail address: fli@kurt.bsd.uchicago.edu (F. Li). www.ics-elsevier.com International Congress Series 1268 (2004) 946 – 951