Reduction of False Positives in a CAD System for GGO Nodule Detection by Means of Neural Classification and CT Coronal View Examination

In this paper, we investigate a procedure for decreasing the number of false positive findings in a reported Computer Aided Diagnosis (CAD) system for the detection of Ground Glass Opacity (GGO) nodules in chest Computed Tomography (CT) images. The proposed procedure consists of two main stages. The first stage is the application of a Radial Basis Function (RBF) neural network on the CT images using a sliding sub-image window scanned over the region of interest (ROI). The RBF network was trained using both, GGO nodule and false positive samples in order to gain the ability to differentiate between both findings. The second stage involves the examination of coronal CT images to confirm the existence of GGO candidates in the transaxial CT images based on the fact that nodular candidates tend to appear similarly in both sections. The algorithm was applied on 2100 slice images containing 27 GGO nodules representing the majority of the typical findings found in the real clinical practice. It succeeded to achieve a detection sensitivity of 96.3% with False Positive (FP) rate of 0.147 FP/slice in case of using the RBF network alone and further improved to 0.06 FP/slice when applying the RBF network together with the coronal images examination, which proves the potential effectiveness of the proposed algorithm.