Gradient Active Contour Driven Lung Segmentation and Lung Nodule Detection System

In this work, we propose novel iterative bounded active contour model initialized by gradient of boundaries to explore uneven lung region from Chest Radiography (CXR) images. This work focused on a unified image transformation framework to exploit the clinically relevant lung regions and retains potentially useful information for texture analyzes and feature extraction and selection to achieve discrimination based on the feature sub set. Moreover this work also focused on availability of gradient information at the lung boundaries active contour model is derived. Finally novel morphological and spatial feature attribues are extracted from ROI segmented lung images for auto classification of CT images. And then optimal sets are evaluated based on the variance measure of the feature values. Experimental results proved that proposed system has brought about a remarkable performance in lung CT image classification. The use of performance and consistency measures has ensured the validity of the experiments. Finally lung nodules detection and multi class classifications are performed with appropriate discrimination measure among normal and abnormal classes using Euclidean distance of hyper planes in SVM classifier. The proposed method is advantageous to early diagnostic process of lung abnormalities and disease. Finally, the performance metrics of proposed feature subsets is compared with state-of-the-art methods and verified using benchmark real-world databases. It is proved that proposed model outperforms competing feature selection methods in terms of accuracy and confusion matrix measures.