We have previously developed a CAD scheme for the detection of lung nodules of both types, solid and GGO nodules. The reported CAD scheme was applied on 715 transaxial CT images containing 25 GGO nodules. It was able to detect 23 nodules and missed 2 nodules achieving detection sensitivity of 92% with False Positive (FP) rate of 0.76 FP/slice.The purpose of this study is to investigate the possibility of further decreasing the number of FP findings reported by our CAD scheme. We will present a proposed neural network that can discriminate GGO nodules efficiently. Also we will benefit from the information revealed from the coronal view examination to reduce the FP rate.Radial Basis Functions (RBF) neural networks are mainly used for pattern classification and they are based on Cover 's theorem on the separability of patterns. This theorem states that nonlinearly separable patterns can be separated linearly if the pattern is cast nonlinearly into a higher dimensional space. Therefore an RBF network converts the input to a higher dimension after which it can be classified using only one layer of neurons with linear activation functions.In the training of the proposed RBF network, we adopted the concept of massive training proposed by Suzuki et al.. This concept, in principle, proposes using a large number of subimages extracted from the input images together with the teacher images containing the distribution for the likelihood of being a nodule. To obtain the output image, a scanning process is performed on the input image with the proposed network and then some score is defined to discriminate the nodular candidates. The algorithm of neural classification is shown in Fig. 1.The second stage of the discriminating procedure is the examination of the coronal CT image coressponding to the location of the assigned GGO nodule in the transaxial sectional image based on the fact that nodular candidates tend to appear in almost circular shapes in both sectional views, i.e. transaxial and coronal sections. Fig. 2 illustrates an example of a GGO nodule in both sectional views.We first apply the neural classification using the proposed RBF network on the transaxial CT images and then examine the corresponding coronal CT images. In our study, we analyzed 2100 transaxial CT images in addition to the corresponding 2142 coronal CT images. Table 1 shows the detection sensitivity and FP rate after applying the proposed discriminating procedure. It is obvious that the combination of both neural classification and coronal view examination improved the experimental results significantly.
RBF Network
Sliding im age window
MemberIn 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 slidin...