Improvement of Overall Performance for Lung Nodule Detection in Chest CT Images

We have developed an automated computerized schema for the detection of lung nodules in 3D CT images obtained by helical CT. In our previous schema, linear discriminant analysis (LDA) and a rule-based method with 53 image features were employed in order to reduce false positives. However, several false positives have remained. Therefore, in this study, we improved the false-positive reduction technique by using the edge image and radial image analysis. Overall performance for the detection of lung nodules was greatly improved. Sensitivity was higher than that of our previous study. Moreover, we evaluated the overall performance of the new scheme by using 69 cases acquired from four hospitals. The average number of false positives was 5.2 per case at a percent sensitivity of 95.8%. Our new scheme would assist in the detection of early lung cancer. (Article in Japanese)