Lung Nodule Detection Using Hybrid Classifier

Lung cancer is one of the most injurious forms of cancer, which is the leading cause of cancer death in many regions of the world. Early detection of lung nodule is essential in reducing life victim. Detection of lung cancer at early stage is not an easy task. Survival rates in lung cancer vary significantly by stage; overall, less than 15% of newly diagnosed patients will survive for 5 years. If patients are diagnosed at the earliest stage, survival rates approach 70% [1]. Cigarette smoking is the most critical reason for lung cancer other factors such as environment pollution, certain chemicals like asbestos, silica, and diesel exhaust cause lung cancer and excessive alcohol may also be consign to lung cancer. The lung nodule detection scheme consists of four steps. They are preprocessing, segmentation, feature extraction and classification. In an existing system, an overlapping nodule identification procedure is designed to help the classification, but this task mainly focused on identifying the nodules located in the intersections among different types. K means clustering and manual analysis method is used for segmentation where the results are not accurate and consumes more time. In short time it is not possible to detect multiple images for cancer detection. Medical images contain a noise which can lead to inaccuracies classification. Extract features from those maximum response filters at eight directions to cover an entire image. Plot histogram for each filter output and concatenate into single histogram. By An abnormality in lung nodule leads to lung cancer which demands an early detection of lung nodule. The study reveals that automatic detection technique of lung nodule with convincing results and increases the speed of analysis. Since the nodules are attached to blood vessels, detection of lung nodule is a challenging task. To deal with this issue MR8 (Maximum response 8) filter bank based approach is used before preprocessing and eight maximum responses are obtained. From that response texture, intensity and gradient features are extracted using LBP (Local Binary Pattern), HOG (Histogram of Oriented Gradient), SIFT (scale invariant feature transform) descriptor respectively. Further the performances of features are analyzed by hybrid classifier. Hybrid classifier approach is to embed SVM (Support Vector Machine) with ID3 (Iterative Dichotomiser).