Pulmonary Nodules 3D Detection on Serial CT Scans

Abstract

This paper describes a Computer-Aided Diagnosis (CAD) system for automatic pulmonary nodules detection on serial CT scans based on shape features. The system recognizes nodules by 3D geometric information through the process of interpolation, segmentation, suspicious area searching and recognition. Firstly, the serial CT images are interpolated to equal scales in X, Y and Z dimensions, in order to recover the original 3D shape of nodules. Secondly, pretreatment is implemented to segment the lung parenchyma region. Thirdly, detect objects called regions of interest (ROIs) as potential nodules by threshold of gray level and region growing. Finally, distinguish ROIs to find real nodules using moment invariants. The experimental results from CT scans data sets demonstrate that the proposed method yields a good performance of nodule detection. The system recognizes all the nodules of the data sets with a reasonable false positive (FP) 1/serial scans.