Automatic Detection and Classification of Pulmonary Nodules on CT Images

Computer-aided detection (CAD) systems are convenient for the automatic lung nodule detection in computed tomographic (CT) images, as the sheer volume of information present in CT datasets is overwhelming for radiologists to process. First, segmentation scheme is used as a preprocessing step for enhancement. Then, the nodule candidates are detected by Eigen value decomposition of hessian matrix and Multi-scale dot enhancement filtering. After the initial detection of nodule candidates using filtering technique, feature descriptors were extracted. The feature descriptor is refined using the process of wall detection and eradication. An Evolutionary Support Vector Machine (ESVM) is trained to classify nodules and non-nodules. The proposed CAD system is validated on Lung Image Database Consortium (LIDC) data. Experimental results show that the detection scheme achieves 98.3% sensitivity with only 11 false positives per scan.