2005 年 125 巻 11 号 p. 1658-1665
This paper aims at developing a Computer Aided Diagnosis (CAD) system used for the detection of pulmonary nodules in chest Computed Tomography (CT) images. These lung nodules include both solid nodules and Ground Glass Opacity (GGO) nodules. In our scheme, we apply Gabor filter on the CT image in order to enhance the detection process. After this we perform some morphological operations including threshold process and labeling to extract all the objects inside the lung area. Then, some feature analysis is used to examine these objects to decide which of them are likely to be potential cancer candidates. Following the feature examination, a template matching between the potential cancer candidates and some Gaussian reference models is performed to determine the similarity between them. The algorithm was applied on 715 slices containing 25 GGO nodules and 82 solid nodules and achieved detection sensitivity of 92% for GGO nodules and 95% for solid nodules with False Positive (FP) rate of 0.75 FP/slice for GGO nodules and 2.32 FP/slice for solid nodules. Finally, we used an Artificial Neural Network (ANN) to reduce the number of FP findings. After using ANN, we were able to reduce the FP rate to 0.25 FP/slice for GGO nodules and 1.62 FP/slice for solid nodules but at the expense of detection sensitivity, which became 84 % for GGO nodules and 91% for solid nodules.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/