Abstract: The computer-aided diagnosis system of lung tumor has great clinical significance for alleviating doctor's pressure and reducing misdiagnosis rate. In this paper, a new method of lung tumor identification based on quantitative three-dimensional shape analysis is proposed. According to the character of CT data and doctor's clinical experience, we defined three two-dimensional shape descriptors correlating with tumor type from different points of view; on this basis, four three-dimensional shape descriptors were defined as input value of radial basis function neural net to realize the automatic classification of lung tumor. The experiment result demonstrates that these shape descriptors can well represent the tumor character:The identification rate is to benign tumor 87.5% and that to malignant tumor is 69.44%.

Key words: three-dimensional shape analysis, radial basis function neural net, lung tumor identification