ICDSIP | March 07-08, 2017 International Journal of Engineering and Advanced Technology (IJEAT) MIT, Aurangabad (Maharashtra) India ISSN: 2249 – 8958, Volume-6, Issue-ICDSIP17, March 2017 76 Published By: Blue Eyes Intelligence Engineering & Sciences Publication Pvt. Ltd. Abstract: Lung nodules which are also called as “Spot on the lung”, a “shadow” or a “coin lesions” are caused by scar tissue, a healed infection or some air irritant but sometimes they are an early sign of lung cancer. The detection of a cancerous lung nodule will facilitate early treatment for lung cancer of patients. Radiologists can detect lung nodules by examining CT scan or X-ray images. Radiologists can be provided with vital information by using automatic lung nodule detection system to assist them in their decision making and treatment suggestion. This paper presents a study of different methods and algorithms used in automatic lung nodule detection. It gives a generalized structure of lung nodule detection that are commonly used in the existing system. It also describes methods and algorithms used for lung nodule detection. The structure includes components which are image acquisition, lung segmentation, nodule candidate detection and segmentation, nodule candidate feature extraction and false positive reduction/classifier. This paper describes methods and algorithms used in every components. Index Terms— Automatic detection, Computed Tomography, Lung Images, Pulmonary Nodule. I. INTRODUCTION A pulmonary nodule is defined as a discrete, well-marginated, rounded opacity smaller than 3 cm in diameter that are completely surrounded by lung parenchyma. Lesions larger than 3 cm are called as pulmonary masses and are more likely to be cancerous than a nodule. There are two types of lung nodules which are benign and malignant. Benign pulmonary nodules are less than 2 cm in diameter and are noncancerous whereas malignant pulmonary nodules are larger than 2 cm and grows constantly and are cancerous. The detection of malignant nodule will facilitate in early treatment of lung cancer. Apart from this pulmonary nodules can also be classified into four classes based on the position of the nodule and surrounding structures [21]. This four classes of lung nodules are as follows. Class 1:- Well-circumscribed, the nodule is present inside the lung region and has no significant connection to surrounding vasculature (vessels). Revised Version Manuscript Received on March 08, 2017. Mahender G. Nakrani, Department of Electronics and Telecommunication Engineering, CSMSS’s Chhatrapati Shahu College of Engineering, Aurangabad, India, E-mail: nakrani.mahender@gamil.com Ganesh S. Sable, Department of Electronics and Telecommunication Engineering, G. S. Mandal’s Maharashtra Institute of Technology, Aurangabad, India, Email: sable.eesa@gmail.com Ulhas B. Shinde, Principal, CSMSS’s Chhatrapati Shahu College of Engineering, Aurangabad, India, E-mail: drshindeulhas@gamil.com Class 2:- Vascularized, the nodule is present inside the lung region and has significant connection to surrounding vasculature (vessels). Class 3:- Pleural tail, the nodule is near the lung walls called pleural surface and connected with thin structure called pleural tail. Class 4:- Juxtapleural, the nodule is significantly connected to the pleural surface. This four class of nodules are shown in fig 1. This paper presents the study on existing methods and algorithms used in automated and semi-automated pulmonary lung nodule detection. It gives a review on existing work and adaptation of different algorithms on lung nodule detection. It also provides generalized structure of lung nodule detection that are commonly used in majority of the published work. The structure consists of 5 components which are Image acquisition, lung segmentation, nodule candidate detection and segmentation, nodule candidate feature extraction and false positives reduction/Classification. In each component various methods and algorithms are employed in different system. This methods and algorithms are reviewed in this paper. II. LUNG NODULE DETECTION METHOD Many automatic and semi-automatic lung nodule detection articles adopts different structure for the method. The most common structure used by many of published work has 6 components in it. Each article involves number of algorithms for a particular component. This structure is displayed in the fig 2. The structure consists of 5 components which are Image acquisition, Lung segmentation, Nodule candidate detection and segmentation, Nodule candidate feature extraction and false positive reduction/Classifier. Many of the articles describes method for all 5 components but few are specific to some components. The articles in which all components are not described utilizes methods which are described for that component from another articles. Few of the article bypasses Lung segmentation component as illustrated in figure 2. The definition of the components and the algorithms and methods existing for that component are as follows. A. Image Acquisition Image acquisition is a process of acquiring image from different imaging modalities for medical images. Lung imaging consists of Ultrasonography, A Review: Automatic and Semi-Automatic Detection of Pulmonary Lung Nodules in Computed Tomography Images Mahender G. Nakrani, Ganesh S. Sable, Ulhas B. Shinde