7 - Lung Cancer: Computer-aided Diagnosis with Computed Tomography

Abstract

This chapter presents the computer-aided diagnosis (CAD) system to detect pulmonary nodules used for lung cancer screening with low-dose helical computed tomography (LDHCT) scanning and discusses the usefulness and pitfalls of the CAD system in mass screening for lung cancer. Lung cancer screening with LDHCT scan was performed on 518 participants (376 men, 142 women; age range 20-85 years, mean age 57.9 years). All 518 procedures were baseline screenings, and no repeat screenings were made. A helical CT scanner was used for this study. The scanning parameters were 120 kilovolt peaks, 50 mA, 10 mm collimation, and 2.0 pitches. Computer-aided diagnosis software includes a detection algorithm for pulmonary nodules and a user interface. The detection algorithm includes a complex segmentation of lung parenchyma, followed by the detection of structures with soft tissue density within the lung parenchyma and a region analysis to evaluate the detected structures in the three-dimensional (3D) data set. All images were obtained at window settings appropriate for lung parenchyma. As for the CAD's performance, the time required to obtain images and diagnose them was ∼7 min. The CAD could not simultaneously provide images from a CT scan when it was running another thoracic examination. Three respiratory physicians separately interpreted all cases according to the General Rule for the Clinical and Pathological Record of Lung Cancer (GRCPRLC) edited by the Japan Lung Cancer Society (JLCS) (2003) (“a,” undetermined, “b,” within normal limit, “c,” old inflammatory lesion, “d,” suspicion of disease other than lung cancer, “e,” suspicion of lung cancer). All three physicians independently judged 46/75 (61.3%) participants as “e,” while one or two physicians judged 29 other participants as “e.”