High-speed Detection of Solitary Nodule in Chest CT Images: Cylindrical Filter and FP Reduction

A number of 2D and 3D detection methods for lung nodules in chest CT images have already been developed. Although, in general, 3D detection methods produce fewer false positives (FPs), they require a huge amount of processing. Therefore, techniques which can reduce computation time are required. In the present study, we investigate a high-speed 3D method for detecting solitary nodules in chest CT images, employing a cylindrical filter kernel that outputs the difference in pixel value between the maximum value of the kernel and the value at the center pixel. Furthermore, we introduce a FP reduction method which uses Linear Discriminant Analysis together with some characteristic shape parameters. Experiments to evaluate the detection performance and computation time revealed that the proposed method is five times faster than existing 3D methods; also, the number of FPs was reduced from 5.11/image to 0.76/image. These results indicate that these techniques are useful for detection of nodules in chest CT images. Keywords— Computer Aided Diagnosis, Lung, Nodule, CT, Image Processing