Thasarathan A/L Gunasegaran and Dr. Haniza Binti Yazid
Abstract: This project is designed specifically to aid radiologists working in imaging department, responsible for making reports regarding the diagnosis result from the CT scan lung images of the patients. In this project, a general Computer Aided Design (CAD) system is developed which aim to detect lung nodules presented in the CT images of possible lung cancer patients. The system comprised of 4 major steps, including pre-processing, segmentation, features extraction and classification. The images that are used in this project are obtained from the LIDC-IDRI database, a large database publicly available. In pre-processing stage, the noises and other artifacts acquired during the scanning process are removed by using suitable filters. In the segmentation stage, an inverse surface adaptive thresholding segmentation method is employed on the images so that the interested nodules in the region of interest (ROI) are segmented. Next, manual cropping is implemented to segment the nodules perfectly in order to extract the features (13 features). Finally, all the extracted features from the segmented nodules are classified by using Back Propagation neural network to test the possible accuracy obtained from the system in segmenting the lung nodules. Based on the result, it is observed that the inverse surface segmentation method works well with CT scan images and segments the interested nodules in the ROI perfectly. The inverse surface method could produce desired level of output segmentation result as comparable to watershed method and Otsu method. The system is successful in classifying the nodules to be nodules or non-nodules with fast computation time, when compared with the conventional method used by the radiologists in current time being.