Computer aided diagnosis system for pulmonary nodules using hierarchical feature extraction

In this paper, we propose a new diagnosis method of pulmonary nodules in CT images to reduce false positive rate (FP) for a high true positive rate (TP) conditions. An essential core of the method is in its hierarchical feature extraction. In the 1st stage, novel orientation features of nodules in a small region of interest (ROI) are extracted in addition to several conventional features, while a more structural feature of a surrounding area of the ROI is extracted in the 2nd stage. Without the orientation features, when TP was 90%, FP was about 65% and 55% in the 1st and 2nd stage, respectively. On the other hand, using the orientation features, FP was about 15% and only 5% in the 1st and 2nd stages, respectively. These improvement of the discrimination rate clearly demonstrates the effectiveness of the proposed hierarchical method on the nodules diagnosis.