A Study on Automatic Detection of Lung Nodules in CT Lung Images K. Varalakshmi a, Dr. S. Ravi b, * b

a1 Department of Computer Science and Engineering, SRM University, Kattankulathur- 603 203 Department of Computer Science, School of Engineering and Technology, Pondicherry University, Pondicherry - 14, India

Abstract Early Detection of cancer may lead to increase the survival rate of lung cancer patients. For the detection of lung cancer CAD system plays a vital role. It consists of four stages called pre-processing or segmentation, nodule detection, feature extraction and classification. There are number of approaches used in the segmentation of lung, nodules and false positive reduction. This paper analysis the approaches used in the segmentation of lung and detection of all possible nodules present in the CT lung image. The main objective of this study is to develop the CAD system that would identify cancer nodules with greater efficiency. Keywords: (Biopsy, Mediastinum, Thresholding, Segmentation, CT Image)

1.

Introduction

includes three type, viz., Adenocarcinoma – Nonsmokers(common in women), Squamous Cell CarcinomaSmokers (Most Commonly in Men), Large Cell Carcinoma. 20-30% of cancers are small cell lung cancer. This includes three more types, viz., Oat Cell, Intermediate and Combined: People in the age 40-60 years are affected by lung cancer. Smoking is the main cause of lung cancer. Other risk factors are genetic factors, tobacco usage for a long term, and occupational risks such as asbestos, arsenic refinery, nickel refinery, chromium exposure, uraninummining,etc., Medical imaging techniques such as X-ray, Computed Tomography (CT), Low-Dose CT (LDCT), Magnetic Resonance Imaging (MRI), High Resolution CT (HRCT), Positron Emission Tomography (PET/CT), Contrast Enhanced CT (CEDT) play a wide role in detection of lung cancer based on the symptoms of patients. Initially, the person is advised to take x-ray to check any abnormal changes are occurred. To be able to see a tumor in an x-ray, it is usually has to be about 1 cm in diameter. At that stage, it will contain 1 billion cells. CT and LDCT help to detect the cancerous nodules in the earlier stage.

Lung cancer is the leading cause of cancer death in developed countries and is rising in alarming rates in developing countries. Lung cancer was not recognized as a disease until 1761. The first link between lung cancer and smoking was reported way back in 1929 by physician Fritz Lickint from Germany. The World Health Organization estimates the worldwide death toll from lung cancer will be 10,000,000 by 2030. The number of global cancer deaths is projected to increase by 45% from 2007 to 2030 (from 7.9 million to 11.5 million deaths). 24 % of men who has the smoking habit have the possibility of getting lung cancer. In India, approximately 85% lung cancer was diagnosed at an advanced stage. With an early detection, almost 40% of all cancer deaths can be prevented. Lung cancer is a disease of uncontrolled cell growth in tissues of the lung. This growth may lead to metastasis, invasion of adjacent tissue and infiltration beyond the lungs. The vast majority of primary lung cancers are carcinomas of the lung, derived from epithelial cells. There are three types of lung cancers, namely, non-small cell cancer, small cell lung cancer and combined. Non-small cell lung cancer constitutes 75-80% of lung cancers. More than 70% of nonsmall cell lung cancers are in stages III and IV. This

* Dr.S Ravi. Tel: +91-9843930392 E-mail:[email protected]

166

Original CT lung Image

Preprocessing / Segmentation Figure 1. CT Lung Image

Feature Extraction

Detection of presence of all nodules is such a challenging task for radiologists. To assist the radiologists, computer-aided diagnosis (CAD) system was developed. CAD system gives the second opinion to the radiologists. Computer aided diagnosis system (CAD) - aims to achieve the accuracy of diagnosis, speed and automation level. It consists of four stages such as preprocessing/segmentation, nodule detection, feature extraction and classification. Preprocessing/Segmentation normally consists in restricting the search space, delimiting the lung, and reducing noises in the image. In nodule detection, the region of the lung is segmented and nodule candidate objects are identified. In feature extraction, among these objects most of the non-nodules are discarded in the false positive reduction stage. In classification, the remaining objects are then classified into nodule and nonnodule. In this paper we summarized only the techniques used in the segmentation algorithms using image processing techniques.

Classification

Cancerous Nodules

NonCancerous Nodules

Figure 2. CAD System for Nodule Detection

1.1 Lung and Nodule Segmentation Lung Segmentation Segmentation is the analysis of images where the object of interest is isolated from the background. In medical image analysis, segmentation is used to delineate specific anatomic structures with the aim of diagnosis of various disorders, locate pathologies, create statistical atlases, quantify structural properties, etc., Thresholding methods, deformable boundaries (active contours and level sets), shape based models, or edge tracking. Lung segmentation is process of removing the background and other parts of the image. The object of interest here is lung region. Lung region is extracted from the original CT image by means of segmentation process. Many researchers work on the segmentation of lung and recorded the performance of segmentation techniques.

Figure 3. Well Circumscribed Nodule

Vascularized Nodule is located centrally in the lung, but has significant vascularization (connection to neighbouring vessels) as shown in figure 4. JuxtaPleural Nodule is a peripheral pulmonary nodule which often exhibits some degree of attachment to the pleural surface (on the external boundary of the thorax) as shown in figure 5.

1.1.1. Lung Nodule Detection After delineating the search space for nodules, detection of nodules is to be done. Detection of lung nodules consists of two functions such as (i) selecting the candidate nodules initially and reducing the false positive. Nodules are low contrast white circular objects within the lung fields. Different shapes of the nodules are: well circumscribed nodule, vascularized nodule, juxtapleural nodule and pleuraltail nodule. Well Circumscribed Nodule is the nodule region N which is distinct from the surrounding lung parenchyma.

Figure 4. Vascularized Nodule

Figure 5. JuxtaPleural Nodule

167

Matthew Brown et al. [9] developed a patient specific model that uses a patient’s base line image data to support in the segmentation of subsequent images so that the nodule measurement was made in regards to changes in size and /or shape of the nodules automatically. The size of the nodule is limited to 5mm and larger in diameter. Shiying Hu et al.[10] developed a fully automatic lung nodule detection. Graylevel thresholding were applied to segment the lung region and left and right lung were separated by identifying anterior and posterior junctions by dynamic programming. Yongbum Lee et al. [11] proposed a novel template matching technique based on the genetic algorithm (GA) template matching for detecting the nodules within the lung area and along the lung wall. Low contrast nodules were missed and found that number of FPs was high as 30/case. Kostis et al.[12] used morphological approach for the detection of nodules. Target is only high contrast nodules and also the nodule is missed if it is attached to the blood vessel. Goo et al. [13] proposed a 3D region growing technique for extracting the nodule candidates based on 3D geometric features. Grey-level thresholding and connected component labeling and background subtraction were applied. In this analysis, they only concentrated on detecting the nodules greater than 5mm in diameter. In [14], 17% of lung nodules can be lost during lung segmentation if the algorithm is not adjusted to the task of nodule detection. Bae et al. [15] developed a computer aided diagnosis (CADx) for high-resolution CT images (HRCT high-resolution computed tomography) using bidimensional and tri-dimensional analysis algorithms. Ge et al.[16] in their work have discussed the major difficulties that should be tackled to detect nodules which are adjacent to anatomical structures such as blood vessels or the chest wall when they have very similar X-ray attenuation and appearance in individual cross-sectional CT images or to detect nodules which are in non-spherical shapes. GadyAgam et al. [17] proposed a correlation-based enhancement filters to enhance the blood vessels, junctions, and nodules and fuzzy shape analysis for vessel tree reconstruction due to low contrast between the lung nodules and blood vessels. Ingrid Slumier et al. [18] proposed a refined segmentation by registration scheme for the detection of low density nodules. Shape model of a normal lung is used to segment the lung. For more severely affected pathological scans, the accuracy achieved is still unsatisfactory. Van Ginneken et al. [21] optimized the Active Shape Model(ASM) to segment the lung regions. ASM is more versatile in medical image applications and better than Active Appearance Model (AAM). Retico et al. [22] proposed a system based on emphasis filters for spherical objects and a neural classification based on voxels of selected regions to reduce false positives. To improve the sensibility of the detection, Li et al. [23] used an emphasis filter in the identification stage and to reduce false positives they used a rule-bases classifier. Jamshid et al. [26] proposed a new region growing algorithm for lung nodule segmentation. This method uses a combination of fuzzy connectivity, distance and intensity

Pleuraltail Nodule is a thin structure of nodule-like attenuation connecting the nodule to the pleural surface.

Figure 6- Pleural tail nodule

2. Literature Survey Hedlund et al. [1] described two computer methods that automatically isolate the lung area of a CT scan. The first method namely, Steep Density gradient boundary estimates the density of large blood vessels, airways and nodules and the second method is called Non-steep density gradient is useful for estimating density and area of the lung parenchyma. He applied edge tracking algorithm for lung segmentation. Manual interaction is needed to provide a seed point to initiate the search to the desired lung- pleural interface. M.L. Giger et al. [2] developed a CAD system and digital image processing techniques were used for the detection of lung nodules. Too much false positive detection was made in radiographs. He observed the evaluation of circularity with incremental thresholding [3] and the evaluation of circularity using a morphological "open" operation. Lo et al. [4] examined the use of neural networks to reduce false positives. thresholding, profile matching analysis and neural network techniques were used to improve the efficiency of CAD system. Only the results were better initially. Jyh-Shyan Lin et al. [5] developed a neural-digital CAD system based on two level convolution neural networks (CNN1 and CNN2) to reduce the false positives. No normalization was applied to the image and no complex feature extraction techniques were involved. Brown et al. [6] presented an automated, knowledge based segmentation routines extract contiguous 3D set of voxels and their feature space representation are posted on the blackboard. For matching the image to model objects based on the features, Fuzzy logic inference engine was used. Manuel G.Penedo et al. [7] developed two level artificial neural network architecture to detect the lung nodules. In the 1st ANN, suspected nodule area (SNA) were found and to reduce the number of false positives, each SNA was transformed to curvature peak space and then introduced to 2nd ANN. Classification process is more complex with many rules regarding the curvature and the compactness and circularity of each point in the SNA. Armato et al. [8, 14] used gray-level thresholding to segment the thorax from the background first and then the lungs from the thorax. A rolling ball filter was further applied to the segmented lung borders to avoid the loss of juxtapleural nodules. The identified lung fields were used to limit the search space for their lung nodule detection framework.

168

information as the growing mechanism and peripheral contrast as the halting criterion. Murphy et al. [30] presented an automatic detection of nodules. Features such as shape index, and curvedness are used in order to detect and 2 successive k-nearest –neighbor classifiers are used to reduce the false positives.

Figure 9. Result of Adaptive Border Marching algorithm [24]

Awai et al.[39] used morphological operators for the detection of nodules. Drawback of this approach is failed to detect the juxtapleural nodules. Kanazawa et al.[38] used clustering technique to identify the nodules. Nodules were missed if it is attached with the lung wall or mediastinum. Kostis et al.[12] used mathematical morphological filtering with various combinations of these basic operators. Greater than 10mm nodules were identified. Lung lobe is a spongy, saclike respiratory compartment in the lung that removes carbon dioxide from the blood and supplies it with oxygen. Left lung consists of two lobes such as upper lobe and lower lobe. Right lung consists of 3 lobes upper lobe, lower lobe, and middle lobe.These lobes are separated by fissure elements called oblique fissures and horizontal fissure is present in the right side to separate the middle lobe from the lower lobe. Li Zhang et al.[19] proposed an automatic lobe segmentation framework with atlas initialization for the lobar fissure segmentation. This system fails to identify the minor fissures separating the upper and middle lobe in the right lung and manual interaction was needed to segment the lobes. This system automatically segments if the complete fissure is present. Wang et al.[20] proposed a method in which major fissures were first segmented in a subset of section, and subsequently a 3D interpolation was applied to obtain the fissure surface. Manual interaction was needed in 2.4% and middle lobe in the right lung was not segmented. Pu et al. [24] set a threshold to initially segment the lung regions. To refine the segmentation and include juxta-pleural nodules, a border marching algorithm was used to march along the lung borders with an adaptive marching step in order to refine convex tracks. Shi et al. [25] used an adaptive shape prior to guiding a deformable model used to segment the lung fields from time-series data. The initial shape was trained from manually marked lung field contours from the population using the principle component analysis (PCA) method and was used to segment the initial time-point images of each subject. Ukil et al.[27] presented an automatic method to segment the lung lobes based on the extracted lobar fissures and the airways on Chest CT scans. 2D ridgeness operation and 2D spline interpolation were used to extract fissures and to extend the incomplete fissures. The later determined the anchor points automatically to ensure the correct

Figure 7 – Result of Adaptive Thresholding [1]

Ye et al. [31] used 3D adaptive fuzzy thresholding to segment the lung region from CT data. Sun et al. [36] presented fully automated approach for segmentation of lungs with such high-density pathologies. 3D robust ASM is used to get a rough initial segmentation of the lung borders and global optimal surface is used for smoothed segmentation of the lungs. This work targets only larger lung nodules and also not optimized for juxtapleural nodules. Bulat et al. [37] presented a novel game-theoretic framework for land-mark based segmentation. It only helps to identify the boundary of the lung region in chest radiographs and object recognition is the future work.

Figure 8. Result of Morphological operations [3]. (a) Original Image b) Binary Image c) Isolated Lungs d) Hollow free lung Mask e) Lung parenchyma

Lo et al. [4] used nodule filtering to suppress the other structures of CT lung image and nodules were separated by simple thresholding. If the nodule is attached with pleural or vessel, that was not detected by this technique. Armato et al.[8] used multiple thresholding technique to separate the lung nodules from the background and other structures was suppressed. Juxtapleural nodule was not detected through his work. Ko et al. [40] used multiple thresholding to segment the nodules from the background. Lee et al. [11] assumed and detected the circular and semicircular candidate nodules using template matching algorithm. Assumptions are not adequate to describe the general geometry of lesions. Pu et al.[24] proposed adaptive border marching algorithm to limit the loss of juxtapleural nodules.

169

boundaries. Then Watershed algorithm is used here for lobe segmentation. Manual interaction is needed if the anchor points didn’t provide enough information. Van rixoort et al.[28] proposed a lobe segmentation method using voxel classification approach. Incomplete fissures may lead to incorrect lobe segmentation. Qiao Wei et al.[29] developed lobe segmentation algorithm for identifying lung lobes from isotropic CT images and selected a threshold to segment the lung regions using histogram analysis. The segmented lungs were then refined using connect-component labeling (CCL) and circular morphology closing. Horizontal fissures cannot be identified under severe pathological cases. JiantoaPu et al. [32] developed fully automated lung lobe segmentation method capable of handling incomplete fissures. Radial Basis Functions (RBFs) was used to extend the individual fissures to the lung boundaries. This won’t be work, if the lung is severely affected by lung diseases. In Pu et al. [33], threshold-based region filling methodology was then used to segment the lung fields as a first step in a pulmonary fissure segmentation framework. For fissure detection , Marching Cube algorithm ,laplacian smoothing and EGI were used. Some plane like structures was incorrectly detected as fissures and small fissures were missed if the lung region was affected severely. Van Rikxoort et al. [34] presented a fully automatic methods for segmentation of the lungs and lobes from thorax CT scans. Region growing approach is used to segment the images and then morphological operations are applied to smoothen the image. Multi-atlas segmentation is applied to detect the errors automatically in the resulting lung segmentation.

a) Original image

c) Lung mask

b) Thresholded image

d) Extracted Lung

Figure 11– Result of Rolling Ball operation [8]

a) Original image

b) Binary Image

c) Lung Mask

d) Lung region

Figure 12– Problem in Morphological operation [3]. Figure 10. Deficiency of thresholding [1] a) Original Image- Similar of intensity between nodule and vessels or bronchi b) Segmented lung Region.

4. Conclusion and future work For the detection of all types of lung nodule, cannot use the same model. Region growing algorithm may not detect the low density nodules. Cluster based techniques are also fails if the lung nodules are having wide range of densities. Redundancy - Overlapping leads to problem in selection of features.(In both region growing and cluster methods, an object detected at higher threshold level belongs to the part of the objects detected at low threshold levels.). Performance level is poor when the dataset is too large. Juxtapleural nodules and vascularized nodules were missed at time of segmentation. Spherical structuring element selected for morphological opening to get the hollow-free lung mask was not working for the entire database of the patient. Processing time overhead and inclusion of unnecessary areas in lung region is increased when reconstructing the lung region using rolling ball algorithm.

3. Discussion Automatic segmentation techniques were used by many researchers in the CAD system. The problems found by the researchers at the time of nodule detection. They are: threshold based segmentation is not accurate and densities (in Hounsfield units) of some pulmonary structures, such as arteries, veins, bronchi, and bronchioles, are very close to densities of the chest tissues. The accuracy of the Shape based segmentation depends strongly on how accurately the prior shape model is registered with respect to CT image. Region growing or flooding algorithms may lead to fusion of neighbouring lobes in the lung lobe segmentation.

170

[16] Z.Y. Ge, B. Sahiner, H. P. Chan, et al. “Computer-aided detection of lung nodules: False positive reduction using a 3D gradient field method and 3D ellipsoid fitting”, Medical Physics, vol.32, pp.2443, 2005. [17] GadyAgam, Samuel G. Armato , and Changhua Wu , Vessel tree reconstruction in thoracic CT scans with application to nodule detection, IEEE Trans. Med. Imaging, vol.24, No.4, pp. 486-499,April 2005. [18] Ingrid Sluimer,MathiasProkop,Bram van Ginneken, Toward Automated Segmentation of the pathological Lung in CT, IEEE Trans. Med. Imaging, vol.24, No.8, pp.1025-1038, August 2005. [19] Li Zhang, Eric A. Hoffman, and Joseph M.Reinhardt, Atlas- Driven Lung Lobe Segmentation in Volumetric X- ray CT images, IEEE Trans. Med. Imaging, vol.25,No.1,pp. 1-16, January 2006. [20] J. Wang , M. Betke, and J.P.Ko, Pulmonary fissure Segmentation on CT, Med.Image Analysis, vol.10,pp.530-547,2006. [21] B. van Ginneken,M.B.Stegman , and M.Loog, Segmentation of anatomical structures in chest radiographs using supervised methods:a comparative study on public database, Medical Image Analysis,vol.10,no.1,pp.19-40,2006. [22] A. Retico, P. Delogu, M.E. Fantacci, I. Gori, A. Preite Martinez, Lung nodule detection in low-dose and thin-slice computed tomography, Computers in Biology and Medicine, 38 (4) 525–534, ISSN: 0010–4825, 2008.doi:10.1016/j.compbiomed.2008.02.001 [23] Q. Li, F. Li, K. Doi, Computerized detection of lung nodules in thinsection CT images by use of selective enhancement filters and an automated rule-based classifier, Academic Radiology 15 (2) ,pp. 165– 175,2008. [24] J. Pu, J. Roos, C. A. Yi, S. Napel, G. D. Rubin, and D. S. Paik, “Adaptive border marching algorithm: automatic lung segmentation on chest CT images,” ComputerizedMedical Imaging and Graphics, vol. 32, no. 6, pp. 452–462, 2008. [25] Y. Shi, F. Qi, Z. Xue et al., “Segmenting lung fields in serial chest radiographs using both population-based and patient-specific shape statistics,” IEEE Transactions on Medical Imaging, vol. 27, no. 4, pp. 481–494, 2008. [26] JamshidDehmeshki, H. Amin, M. Valdivieso, and X. Ye, “Segmentation of pulmonary nodules in thoracic CT scans: a region growing approach,” IEEE Transactions on Medical Imaging, vol. 27, no. 4, pp. 467–480, 2008 [27] SoumikUkil and Joseph M.Reinhardt , Anatomy -Guided Lung Lobe Segmentation in X-ray CT images, IEEE Trans. Med. Imaging, vol.28,No.2,pp. 202-214, Feb 2009. [28] E.M.VanRikxoort, B.debHoop, S.Van de Vorst, M.Prokop, and B. van Ginneken,AutomaticSegmetation of Pulmonary segments from volumetric chest CT scans, IEEE Trans. Med. Imaging,vol.28,no.4,pp.621-630, Apr.2009. [29] QiaoWei, YaopingHu,GaryGelfand, and John H. MacGregor, Segmentation of lung lobes in high resolution isotropic CT images,IEEE Trans. Med. Imaging, vol.56, No.5, pp.1383-1393, May 2009. [30] K.Murphy, B.vanGinneken, A.M.R Schilham, B.J. de Hoop, H.A. Gietema, M.Prokop, A large- Scale evaluation of automatic pulmonary nodule detection in Chest CT using local image features and knearest- neighbor classification,Med. Img. Analysis, 13,757-770, 2009. [31] X. Ye, X. Lin, J. Dehmeshki, G. Slabaugh, and G. Beddoe, “Shapebased computer-aided detection of lung nodules in thoracic CT images,” IEEE Transactions on Biomedical Engineering, vol. 56, no. 7, pp. 1810–1820, 2009. [32] J. Pu, J. K. Leader, B. Zheng et al., “A computational geometry approach to automated pulmonary fissure segmentation in CT examinations,” IEEE Transactions on Medical Imaging, vol. 28, no. 5, pp. 710–719, 2009. [33] JiantaoPu,BinZheng,Joesph K. Leader,Carl Fuhrman, Friedrich Knollmaan,AmyKlym, and David Gur, Pulmonary lobe segmentation in CT Examinations Using Implicit Surface Fitting, IEEE Trans. Med. Imaging, vol.28,No.12, pp.1986-1996, Dec 2009. [34] Eva M. Van Rikxoort, Mathias prokop,Bartjan,de hoop, Max.A,Viergever, JosienP.W.Pluim, and Bram van Ginneken, Automatic Segmentation of Pulmonary Lobes robust against Incomplete Fissures, IEEE Trans. Med. Imaging, vol.29,no.6, June 2010.

As of now the researchers faced the difficulty of identifying the nodules which are attached to the lung wall, mediastinum or blood vessels. This is because the segmentation of lung region was not accurately done. To improve the accuracy of segmentation of lung, in future the adaptive border marching will be enhanced for proper detection of lung nodules. Adaptive Border Marching (ABM) smoothes the lung border and will be more helpful to identify missed juxtapleural nodules and also minimizes under segmentation and over segmentation of non-lung portions.

References [1] L. W. Hedlund, R. F. Anderson, P. L. Goulding, J. W. Beck, E. L. Effmann, and C. E. Putman, “Two methods for isolating the lung area of a CT scan for density information,” Radiology, vol. 144, pp. 353– 357, 1982. [2]Giger ML, Doi K, MacMahon H: Image feature analysis and computeraided diagnosis in digital radiography. 3. Automated detection of nodules in peripheral lung field. Med Phys. 15:158-166, 1988. [3] Giger ML, Ahn N, Doi K, et al: Computerized detection of pulmonary nodules in digital chest images: Use of morphological filters in reducing false-positive detections, Med Phys 17:861-865, 1990. [4] S. C. Lo, M. T. Freedman, J. S. Lin, and S. K. Mun, “Automatic lung nodule detection using profile matching and back propagation neural network techniques,” Journal of Digital Imaging, vol. 6, no. 1, pp. 48– 54, 1993. [5] Jyh- Shyan Lin, Shih- Chung, B. Lo, Akira Hasegawa,Matthew T. Freedman, and SeongK.Mun, reduction of false positives in Lung Nodule detection using a Two-Level Neural classification, IEEE Trans. Med. Imaging. Vol.15,No.2, pp. 206-217, April 1996. [6] Matthew S. Brown , Michael F.McNitt-Gray, Nicholas J. Mankovich, Jonathan G. Goldin, John Hiller, Lawrence S. Wilson and Denise R. Aberle, Method for segmenting Chest CT image data using an Anatomical Model: prelimnary results, IEEE trans. Med. Imaging, vol.16, No.6, Dec 1997. [7] Manuel G.Penedo, Maria J. Carreira,AntonioMosquera, and Diego Cabello, Computer-aided diagnosis: A neural network based approach to lung nodule detection, IEEE Trans. Med. Imaging. vol.17 No.6,pp. 872-880, Dec 1998. [8] S.G. Armato, M.L. Giger, C.J. Moran, J.T. Blackburn, K. Doi, H. MacMahon, Computerized detection of pulmonary nodules on CT scans, Radiographics 19 (5) (1303–1311,1999. [9] Brown MS, McNitt-Gray MF, Goldin JG, Suh RD, Sayre JW, Aberle DR. Patient-specific models for lung nodule detection and surveillance in CT images. IEEE Trans. Med. Imaging. 2001;20:1242–1250. [10] Shiying Hu, Eric A. Hoffman, and Joseph M.Reinhardt,Automatic lung segmentation for accurate quantitation of volumetric x-ray CT images, IEEE Trans. Med. Imaging, vol.20,No.6, pp.490-498, June 2001. [11] Yongbum Lee, Takeshi Hara, Hiroshi Fujitha , Shigeki Itoh , and Takeo Ishigaki, Automated detection of pulmonary Nodules in Helical CT images based on an improved template- matching technique, IEEE Trans. Med. Imaging. Vol.20, No.7, July 2001. [12] W. J. Kostis, A.P.Reeves,D.F. Yankelevitz, and C.I Henschke, Three Dimensional segmentation and growth rate estimation of small pulmonary nodules in helical CT images, IEEE Trans. Med. Imaging, vol.22,no. 10,pp.1259-1274,Oct.2003. [13] Jin Mo Goo,Jeong Won Lee, Hyun Ju lee, Seunghwan Kim, Jong Hyo Kim, Jung – Gi lm, Automated Lung Nodule Detection at LowDose CT: preliminary Experience,Korean J radiology 4(4),Dec 2003. [14] S.G. Armato, W.F. Sensakovic, Automated lung segmentation for thoracic CT: impact on computer-aided diagnosis, Academic Radiology 11 pp.1011–1021.,2004. [15] K.T. Bae, J.-S. Kim, Y.-H. Na, K.G. Kim, J.-H. Kim, Pulmonary nodules at chest CT: effect of computer-aided diagnosis on radiologists detection performance, Radiology 236,pp. 286–293,2005.

171

[35] E.M.vanRikxoort and B.vanGinneken,Automatic Segmentation of the lungs and lobes from thoracic CT scans,Diagnostic image Analysis group,Department of Radiology,Netherlands,2011. [36] S.Sun,C.Bauer,andR.Beichel,Automated 3D Segmentation of lungs with lung cancer in CT data using a novel robust active shape model approach, IEEE Trans. Med. Imaging, vol.31,no.2,pp.449-460,Feb 2012. [37] BulatIbragimov, BostjanLikar, FranjoPernus, and TomazVrtovec, A Game- Theoretic Framework for landmark based image segmentation, IEEE trans. Med. Imaging,, vol.34, no.9, pp. 1761-1776, Sep 2012.

[38] K.Kanazawa , Y.Kawata, N.Niki, computer- aided diagnosis for pulmonary nodules based on helical CT images , Computerized Medical Imaging and Graphics,vol.22, no.2, pp.157-167,1998. [39] K.Awai,K. Murao, A.Ozawa , Pulmonary nodules at chest CT: effect of computer –aided diagnosis on radiologists’ detection performance, Radiology, vol.230,no.2,pp.347-352,2004. [40] J.P. Ko and M. Betke, Chest CT: automated nodule detection and assessment of change over time – preliminary experience, Radiology, vol.218, no.1,pp.267-273,2001.

172