Abstract: Computer–aided detection (CAD) can help the radiologists to detect pulmonary nodules at an early stage. In this paper we propose a CAD system to detect the pulmonary nodules from the segmented lungs from Computer Tomography images and classify the abnormalities. First the lung is segmented from the CT images using Watershed Transform then the nodules are extracted. Here for detection of lung nodules some features are considered such as 1) Shape features: Shape based feature descriptor are used. The shape of the nodules is considered for detection of nodules. 2) Texture features: 2D Haar Wavelet Transformation is used which is used to obtain the texture information in different levels of decomposition. 3) Intensity features: The intensity of the nodules is taken as a feature which helps us in putting a threshold which is taken on trial and error manner which help us in detecting the nodules. 4) Context features: Specifies the location of nodule. Classification is done using three different classifiers like SVM, ANN & KNN to increase the efficiency and decrease the error rate. Finally a comparative study is made with respect to error rate, efficiency and training ratio.
Keywords: Lung cancer, Image pre-processing, Image segmentation, Feature Extraction, Classification.