ABSTRACT

Lung cancer continues to rank as the leading cause of cancer deaths among Americans. The number of lung cancer deaths each year is greater than the combined number of breast, colon, and prostate cancer deaths. Evidence suggests that early detection of lung cancer may allow more timely therapeutic intervention and thus a more favorable prognosis for the patient (Heelan et al. 1984; Sone et al. 1998). Accordingly, lung cancer screening programs are being conducted in the United States (Henschke et al. 1999; Miettinen and Henschke 2001; Henschke et al. 2001; Swensen et al. 2003), Japan (Kaneko et al. 1996; Sone et al. 1998; Sone et al. 2001; Nawa et al. 2002), and other countries with low-dose helical computed tomography (CT) as the screening modality. Helical CT, however, generates a

CONTENTS

13.1 Introduction ........................................................................................................................ 297 13.2 Databases ............................................................................................................................ 299

13.2.1 Database of Low-Dose CT Images ....................................................................... 299 13.2.2 Database of Chest Radiographs ........................................................................... 301

13.3 CAD Scheme for Thoracic CT .......................................................................................... 301 13.3.1 Current Scheme for Lung Nodule Detection in Low-Dose CT ....................... 301 13.3.2 Architecture of Massive Training ANNs for FP Reduction ............................302 13.3.3 Training Method of Expert MTANNs ................................................................304 13.3.4 Scoring Method for Combining Output Pixels .................................................305 13.3.5 Mixing ANN for Combining Expert MTANNs ................................................306