Heuristic Search Based Parameterized Level Set for Automated Lung Parenchyma Segmentation and Nodule Extraction

Abstract

Image Segmentation is a very important image processing technique now a days. It is greatly used in the sector of medical image processing. Segmentation for lung areas from CT images is important task on understanding tissue construction, computing and extracting abnormal areas as well as parenchyma segmentation. There are many application of the lung image segmentation in lung parenchyma segmentation, lung nodule extraction, lung tumor classification, lung cancer detection and so on. These segmentation techniques, some of them are semi-automatic and some of them are fully-automatic. Some fullyautomatic techniques include thresholding, snakes, level set, region growing, bayesian network, hierarchical multi-scale, gradient descent and so on. The main objectives of the lung image segmentation are the efficiently and accurately segment the lung parenchyma and the lung nodule. The above mentioned techniques have their strength in their own dataset to segment correctly but they cannot perform well in all kinds of dataset. Another thing which are very important here to reduce time complexity, memory space and calculation complexity. In this paper we proposed a method which use the bi-directional chain method to select the seed points near the lung parenchyma automatically. It helps our next step of the proposed method- parameterized level set method. ii For level set method it is necessary to select random points in the image through which it converges to the boundary of the object(s). But in our proposed method we do not use the random seed points. Rather we use the particular seed points near the lung parenchyma got from bi-directional chain method in which we use uninformed heuristic search and memoization technique. Then we implement the level set method to get the lung parenchyma correctly in reduced computational time. It perfectly segment out the lung parenchyma along with any irregular boundary and abnormal shape. Next step of our proposed method is to segment out the lung nodule accurately even if it resides near the boundary and having any abnormal shape. Experiment is performed employing 60 CT image sets from 18 patients and satisfactory results are obtained. Obtained results are shown along with a discussion.