An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review

Full text

An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review
An Automated System for the Detection of Lung
Cancer in CT data at Early Stages: Review
Satya Prakash Sahu1, Narendra D. Londhe2 and Shrish Verma3
1 Department of Information Technology National Institute of Technology Raipur, India, Email: spsahu.it@nitrr.ac.in
2 Department of Electrical Engineering National Institute of Technology Raipur, India, Email: nlondhe.ele@nitrr.ac.in
3 Department of Electronics & Telecommunication Engineering National Institute of Technology Raipur, India, Email:
shrishverma@nitrr.ac.in
Abstract: The severity of lung cancer presented the challenging tasks to the researchers for the development of
efficient automated systems or Computer-aided diagnosis (CAD) systems to detect of lung cancer in primary stages.
Due to this reason, automated systems for the detection of lung cancer have been explored in large number of research
articles. This paper emphasizes the general CAD systems for the diagnosis of lung cancer which consist of major
steps like preprocessing, lung segmentation, nodule detection in the lung area, nodule segmentation and analysis
based on volume & shape, growth, texture & appearance followed by classification as malignant or benign and
diagnosis. This paper focuses the current technical issues, existing methodologies, various referred databases and
validation with ground truths and finally the description of achieved performance and its analysis & comparisons
with some popular metrics.
Keywords: Computer aided diagnosis (CAD) systems, Medical Image Processing, Lung cancer, Lung segmentation,
Nodule Detection and Computed Tomography.
1. INTRODUCTION
According to the statistical analysis of American Cancer Society [1] in 2016, there are 224,390 new cases of
lung cancer in the United States, and deaths from lung cancer are estimates around 158,080. Among all kind of
cancer, lung cancer is the leading cause of cancer death which includes both men and women, i.e. lung cancer
causes about one cancer death out of four. The total of colon, breast and prostate cancer are less than the Lung
cancer each year. The survival rate of lung cancer is five year, which is very poor because the diagnosis of lung
cancer is usually done in later stages. Earlier the evaluation of CT scans was generally done through manually
resulting very exhaustive tasks. So the clinical identification and handling of diseases in managed way are
required that would lead to enhancement of survival rates [2]. Thus a computer-aided diagnosis (CAD) system
can be extremely useful to reduce the manual overhead and to detect the lung cancer in early stages so that it can
improve the survival rate of lung cancer patient and provide the better healthcare services in this field. These

Satya Prakash Sahu, Narendra D. Londhe and Shrish Verma
facts attracted the attention of many researchers to develop an effective and efficient automated CAD system
that involves steps of image processing into the patient’s CT image to detect the cancerous cell and do the
required analysis to diagnose [3].
The various common non invasive imaging modalities are computed tomography (CT), contrast enhanced
computed tomography (CE-CT), low-dose computed tomography (LDCT) and Positron emission tomography
(PET) for detection and diagnosis of lung cancerous cell. The block diagram of General CAD system for the
detection of lung cancer is as shown in Figure 1. The input to the CAD systems is the image acquired by specific
imaging modalities. The preprocessing step may contain transformation and DICOM specific operations, format
and window size identification finally thorax part extraction. The lung segmentation is crucial and essential
step; it basically extracts the portion of lungs from CT chest image for finding the accurate ROIs. The next step
is to detect and segment the nodules in segmented lungs and finally the detected nodules are classified into
malignant (cancerous) and benign (non-cancerous) on the basis of shape, growth and texture & appearance
analysis followed by diagnosis.
Figure 1: General CAD System for Lung Cancer

An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review
In CADe systems, a system for lung nodules detection comprises image acquisition, preprocessing, lung
segmentation, nodule segmentation & detection of segmented candidates into nodules and non-nodules (normal
components like vascular organs, etc.). However in CADx systems, the lung nodules detected in CADe systems
or by radiologists are identified whether these are malignant or benign.
In the remaining section of paper, the different processing steps of CAD systems are discussed such as
lung segmentation and its various investigations by researchers, nodule segmentation, detection and classification
with some performance metrics.
2. LUNG SEGMENTATION
The lung segmentation is an essential and preprocessing step that reduces the search space and decreases the
overhead for further process i.e. detection of the nodules. It is process of extraction of the lungs in CT chest
images and separates the ROIs from muscles, fats, and other attached pulmonary structures like veins, arteries,
bronchi, and bronchioles. The accuracy of segmentation is very important and well stated by S. G. Armato and
W. F. Sensakovic that due to poor and inaccurate lung segmentation approximately 4.9 % to 17% true lung
nodules are not detected and missed [2]. Though, the presence of noise, low contrast and intensity inhomogeneity
and in CT images makes the lung segmentation a difficult task but the most challenging problem for researcher
is to recognize the nodules attached to pleural surface (i.e. juxtapleural) and to include these regions in the
segmented lungs. Numerous algorithms have been anticipated in the literature, such as thresholding based
segmentation, region growing based segmentation, segmentation based on fuzzy c-means (FCM) clustering
approach, active contour models based segmentation, Intensity-based segmentation, Level set image segmentation,
Texture-based image segmentation, and so forth. The widely used lung segmentation techniques are:
Active contour based Segmentation: Minimization of the energy function in each iteration using some
dynamic contour is the central idea behind active contour model (ACM). The problem in accurate segmentation
is due to lack of organ tissue homogeneity in texture and shape of various slices of CT image. Initialization and
poor convergence related problem also limit the effectiveness of active contour method, which has been addressed
in many research studies [5, 6 and 7]. One of the approaches for convergence has been given by Laurent D
Cohen [6], to the object’s boundary, an external force will apply to guide the snake which is similar like inflates
or deflates a balloon. When snake initialized far from the desired boundary then the convergence is improved,
but weak edges boundary may not be detected if the strength of the balloon force is too high. The introduction of
Gradient vector flow (GVF) as an external force has been given as an important solution by Chenyang Xu and
Jerry L Prince [5] where both the problems initialization and convergence has been addressed. In this direction
further improvement has been done by Wang et al. [8] with approach called normally biased GVF (NBGVF) as
an external force. This method keeps existing desired property of GVF with weak edges and smoothen the noise.
The related work for improving the convergence has been also addressed in [10, 11, and 14]. Another reason for
the poor convergence is the smaller searching space. For the initialization issue Chan et al.[11] adopted a ACM
that exploit techniques of curve evolution. Mumford–Shah function [5] skipped the need of precise boundary
and doesn’t even require for smoothening of the initial image to detect the object. Cui et al.[12] defined energy
function to draw contour toward object boundaries with local intensity distribution for flexible initialization and
effective noise handling. Athertya et al.[13] has shown fully automatic method for ACM that avoids the human
interaction.
Thresholding Based segmentation: Lungs portion are darker as compared to other anatomical structure in
the thorax region so with the selection of optimum threshold the segmentation of lungs have been utilized in
number of research studies. Hu et al.[15] adopted the iterative threshold with some morphological operations.
Gao et al.[16] proposed the threshold based approach with some preprocessing, region growing and morphological
smoothing. Wei et al.[17] used threshold to segment the lung region using histogram analysis, Ye et al.[18] used
adaptive fuzzy thresholding for the lung segmentation from CT data. Baniani et al. [19] has given the approach

Satya Prakash Sahu, Narendra D. Londhe and Shrish Verma
Table 1
Current methodologies review for lung segmentation
Authors, Year Technique/Method Database & Image Size Performance Ground Truth
Dimension Accuracy
Hu et al., 2001 [15] Iterative threshold, 24 dataset from 8 512x512, 3 mm thin RmsD = 0.54 mm 229 manual traced
dynamic programm- Subjects, 3D (0.8 pixel) images
ing, morphological
Operations
Yim et al., 2005 [34] Region growing, 10 subjects, 3D 512x512, 0.75 - 2.0 RmsD =1.2 pixel 10 manual traced
connected com- mm thin data
ponent
Gao et al., 2007 [16] Thresholding 8 subjects, 2D 512x512x240 DSC=0.9946 8 manual traced
datasets
El-Baz et al., 2008 statistical MGRF 10 image datasets, 512x512x182, 2.5 Segmentation 1820 manual traced
[35] model 3D mm thin Accuracy = 0.968 images
Kockelkorn et al., Prior training, 22 scans, 3D 512x512, 0.9 - 1.0 OM=0.96AD= 12 manual traced
2010 [28] statistical classifier mm thin 1.68mm data
Sofka et al., 2011 Active Shape Based 260 scans, 3D 512x512, 0.5 - 5.0 SCD = 1.95 68 manual traced
[30] mm thin data
Hua et al., 2011 [36] Graph Search 15 scans, 3D 512x512, 0.3 - 0.9 HD=13.3 pixel 12 semi-automated
mm thin Sen.=0.986 traced data
Spec.= 0.995
Nunzio et al., 2011 Threshold, Region 130 HRCT Scans, 512x512, 1.25 mm OM=0.96±0.02, 36 manual traced
[37] Growing, Morpho- 3D thin AD=0.74±0.05, data
logical operation RmsD =0.57±0.04
Pu et al., 2011 [31] Threshold & Geo- 230 scans, 2D & NA RmsD=0.15±0.092 20 manual traced
metric modeling, 3D Max error =7.82± data
shape based analysis 3.37
OM=95.1+2.0%
Sun et al., 2012 [32] Active shape based 30 MDCT scans, 512x512x424-642, DSC=0.975 30 manually
3D 0.6-0.7 mm thin AD=0.84 mm corrected traced
SPD=0.59 mm data
HD=20.13 mm
Abdollahi et al., statistical MGRF 11 scans, 3D 512x512x390, 2.5 DSC =0.960 11 manual traced
2012 [38] model mm thin data
Cortez et al., 2013 3D region growing 11 subjects, 3D 512x512, 0.5 - 2.0 2 pulmonologists 11 manual traced
[39] & Threshold mm thin qualitatively eval- data
uated seg. results
64.78 % & 68.18%
satisfactory
Keshani et al., 2013 Adaptive Fuzzy 63 subjects (4 512x512, 0.625 - Segmentation 8 manual traced
[40] Thresholding, Active groups: 4, 4, 5 & 50 5.0 mm thin Accuracy = 0.981 data I & II group
contour subjects of LIDC ), and all annotations
3D provided from III
& LIDC
Kuruvilla et al., Otsu Threshold, 155 subjects, 3D 512x512, 0.75-1.25 RmsD =0.0942 LIDC annotated
2014 [41] Morphological mm thin and manually
operation traced data
Orkisz et al., 2014 Threshold, Mor- 20 scans, 3D 512x512, 0.6-1.0 differentiated ves- 20 manual traced
[42] phological operation mm thin sels from bronchial data

An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review
to determine single or more threshold value based on histogram. An optimal multilevel thresholding has been
proposed by Maitra et al. [21] an enhanced particle swarm optimization (PSO) variant, the approach decomposed
multidimensional swarm into numerous one-dimensional swarms, to estimate overall fitness swarms exchange
information among themselves. Multilevel thresholding method termed as maximum entropy based artificial
bee colony thresholding (MEABCT) anticipated by Ming-HuwiHorng [20], for selecting the passable thresholds
this method simulates the behavior of Artificial bee colony algorithm. Xu et al. [22] proposed an improved
discrete quantum particle swarm optimization (IDQPSO) algorithm, based on 2D threshold particle swarm
binary-encoded method for accelerating the converging and local searching. Otsu thresholding technique is one
of the frequently used thresholding methods, but is having the drawback of complexity in computation and
processing time. Otsu adaptive thresholding segmentation has been proposed by Kim et al. [23] based on bimodal
histogram which inspect the bimodality of each region and shown the better performance. Helen et al. [24]
improved the performance of 2D Otsu-based thresholding segmentation using PSO. Banimelhem et al. [25]
proposed memetic algorithm (MA) for image segmentation in order to speed up the searching process and
generating faster solution.
Shape based segmentation: This technique uses the prior shape information of lungs. Some previous shape
parameters like edges and points are analyzed and used to formulate the variational energy framework for
segmentation. This energy framework guide and help to deformable models for the segmentation of lung fields.
Shi et al.[26] proposed the population based and patient specific shape statistics to constraints the deformable
model. The earlier segmentation results and shape statistics have been collected online and updated when some
new segmentation results available and thereby used to refine and improve the segmentation accuracy. Annangi
et al.[27] used prior shape and some low level features for lung segmentation in x-ray images. Kockelkorn
proposed user interactive techniques for lung segmentation in CT image with severe abnormalities where prior
& region growing walls spec. = 0.848
Zhou et al., 2014 Preprocessing, 20 MDCT scans, 512x512x210-540, OM=95.81±0.89% 20 manual traced
[43] FCM and adaptive 3D 0.6-1.0 mm thin AD=0.63±0.09mm data
thresholding
Shen et al., 2015 Preprocessing, 233 scans, 3D 512x512 Re-inclusion rate= 10 manually traced
[44] adaptive Threshold 92.6%, over seg = data
0.3% and under
seg=2.4%
Wang et al., 2016 Principal compon- 45 scans, 3D 512×512×275 - RmsD=1.6957± 45 manually traced
[45] ent, Morphological 512×512×502, 0.6568 mm , data
op, Connected 0.55-1.0mm. thin AD= 0.7917±
reigion based and 0.2714 mm
contour segmentation VOE=3.5057±
1.3719mm
VD=11.15±69.63
cm3
Abbreviations: DSC - Dice similarity coefficient; OM - overlap measure = TP/(TP + FP + FN); Sen. - sensitivity = TP/(TP + FN).
Spec- specificity = TN/(TN + FP); RmsD - root mean square difference = distance between the segmentation and the ground truth;
AD - mean absolute surface distance; HD - Hausdorff distance = mean maximum distance of a set to the nearest point in the other
set; SPD -mean signed border positioning error; SCD - symmetrical point-to-mesh comparison error; VOE – volume overlap error;
VD – volume difference,
LIDC – Lung Imaging Database Consortium (https://imaging.cancer.gov/programsandresources/informationsystems/lidc)
(contd...Table 1)
Authors, Year Technique/Method Database & Image Size Performance Ground Truth
Dimension Accuracy

Satya Prakash Sahu, Narendra D. Londhe and Shrish Verma
shape are trained using k-nearest-neighbor classifier and on the basis of classification results can be corrected.
Kumar et al.[29] presented two stage approach for automatic lung lobes segmentation where fissure region are
found then fissure location and curvatures are identified within these regions. Sofka et al. [30] combined statistical
shape model with anatomical information based pattern recognition technique for the robust lung segmentation.
Pu et al. [31] proposed an approach with geometric modeling and shape based “break-and-repair” strategy for
segmentation of a medical image. Sun et al. [32] proposed novel approach robust active shape model to roughly
segment the outline of lungs, then optimal surface finding method is used to further adapt the initial segmentation
result. Gill et al. [33] given the feature based atlas approach for initialization of active shape models for
segmentation.
The research studies of some current techniques in lung segmentation are reviewed in Table-1 along their
technical issues and effectiveness with different data sets and modalities.
3. NODULE SEGMENTATION AND CLASSIFICATION
The lung nodules are white spherical (circular) matter having lower contrast found in lung region. The most
challenging tasks for the segmentation of such nodules are their unfavorable locations i.e. if attached to wall of
parenchyma called as “juxtapleural nodules”; nodules attached to vessels of blood called as “juxtavascular
nodules”. The other type of nodules which are sub-solid in nature and usually having lower HU values as
compared to other nodules are called as “ground-glass opaque” (GGO) nodules. Also some nodules are small in
size but having critical role for the prior detection of cancer in lung region. For the juxtapleural cases, a number
of research studies have been addressed the solution and the most common method used is “pleural-surface-
removal” (PSR) given by [46, 47, 49, 50, 51, 55]. This method can be implemented in global scenario where
firstly complete segmentation of lung from CT slices then the outcome is adopted as negative mask to avoid the
non-targeted wall portion to be considered in segmentation results. For the juxtavascular cases of nodules the
most popular method given by [46-49, 55] is morphological operations i.e. filtering based on erosion, dilation,
opening, etc. GGO nodules offer the challenging task to draw the exact boundaries and modeling of the irregular
contours. The widely used method addressed by [48, 52, 53] is based on probabilistic classification of voxels.
For dealing with small nodules the common method called “partial-volume-effect” (PVE) and its variants given
by Ko et al.[54] and Kuhnigk et al.[55] respectively. Summarizing about the nodules all authors cited here have
their opinion that juxtapleural and sub-solid nodules are very hard to characterize and offered most challenging
task for accurate segmentation.
The classification of nodules to be malignant or benign in an automated systems the common approach
followed by various authors in number of research studies are based on domain specific features. The most
common features considered for lung nodules are depend upon the nodules shape, appearance, textures, etc. The
steps followed for the classification are: i) selection and extraction of useful features, ii) extracted features are
organized, analyzed and processed by some specific classifier algorithms e.g. SVM, LDA, ANN, etc., iii) classifier
algorithms are designed and implemented based on features, iv) sampled data sets are trained (iteratively with
some defined limiting conditions) using classifier algorithm considering the extracted important features and
the outcome are validated with some benchmarks or ground truths, v) finally in testing phase entire data sets are
processed and nodules are classified into malignant or benign and further the results are observed with ROC
analysis and subsequently criteria may be followed for the performance enhancement i.e. reduction of false
positives. Kwata et al.[56] used shape based features of 3D nodules such as surface curvature and ridge line. El-
Baz et al. [57] proposed a novel approach based on 2D visual appearance features where HU values of malignant
nodule appearance were modeled with rotational invariant second order Markov-Gibbs random field (MGRF).
Han et al. [58] investigated the 2D texture features like linear binary pattern (LBP), haralick and gabor and
further extended these features to 3D spaces and observed that haralick feature achieved greatest area under
curve (AUC) values. Recently for the irregularities in nodules shape Dhara et al. [59] had given diagnostic
parameters like sphericity, spiculation and lobulation.

An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review
Table 2
Review of current methodologies for nodule segmentation and classification
Authors, Year Technique/Method Features Used Database Performance Clinical Validation
classification Accuracy
Antonelli et al., RFCM, Region 3D geometrical 30 scans Sen.= 0.925 10 manual traced
2011 [60] growing, Threshold- features: volume, Spec.= 0.835 datasets
ing, Multiclassifier sphericity, radius,
systems aggregated max compactness,
from statistical, NN max circularity and
and decision tree. max eccentricity
Keshani et al., 2013 Adaptive Fuzzy 2D stochastic and 63 subjects (4 Detection rate= 89%, 8 manual traced
[40] Thresholding, 3D anatomical groups: 4, 4, 5 & 50 Fp/scan = 7.3 data I & II group
Active contour, features subjects of LIDC) and all annotations
classification by provided from III
SVM & LIDC and
ANODE09
challenge
Filho et al., 2014 Threshold & region Shape features: 800 exams (640 for Sens = 85.91%, LIDC-IDRI
[61] growing and SVM spherical dispro- validating and 160 Spec = 97.70% ,
portion, density, for testing) accu = 97.55%,
sphericity, weighted
radial distance and
radial shape index
Choi et al., 2014 Threshold, 3 3D shape based 84 CT scans Sens = 85.91%, Manual traced
[62] D connected com- features Fp/scan = 6.76 data; LIDC
ponent and SVM datasets
Kuruvilla et al., Otsu Threshold, Statistical features 155 patients Train. Func 1: LIDC datasets
2014 [41] Morphological Accu = 91.1 %,
operation, Spec =100%,
Statistical method: Sens=91.4% ,
mean, SD, skewness, MSE = 0.998
kurtosis, fifth & Train. Func 2:
sixth central Accu =93.3 %,
moment; feed MSE = 0.0942
forward BPN
Alilou et al., 2014 Region growing, 3D features: volume, 60 CT scans Sens = 0.80 LIDC datasets
[63] Morphological min dimension size, Fp/Scan= 3.9
operation, Rule max dimension size,
based & SVM eccentricity, com-
classifiers pactness, min
intensity, dist to
center2D features:
area, circularity,
intensity, dist to
center, max & min
intensity
Tasci et al., 2015 Morphological Shape and texture 24 CT scans With 33 features LIDC datasets
[64] operation, Thres- based features nodule recog AUC
hold, generalized value = 0.9679
linear model
regression (GLMR)
Farahani et al., 2015 Region growing, Roundness, 60 CT Scans MLP – Accu = LIDC datasets

Satya Prakash Sahu, Narendra D. Londhe and Shrish Verma
The research studies of some current trends used by various authors in nodule segmentation and classification
are reviewed in Table-2.
4. DISCUSSION AND CONCLUSIONS
Development of effective and efficient automated (CAD) systems that can detect the cancerous matter or nodules
in primary stage is very useful and important because early detection and diagnosis may lead to the better and
appropriate treatments and thereby decrease the mortality from the lung cancer and enhance our healthcare
services too. In this paper we have addressed 68 publications on various phases of automated systems. The
recent trends, technological aspects and methodologies have been focused with their limitations and strengths.
In the major phase of lung segmentation the technical issues and challenges have been discussed with number of
research articles and the common techniques are categorized such as threshold, active contour, prior shape
model, etc. based approaches are reviewed with their proper validations. In nodule segmentation and classification
phase the complexities related to various nodule types are discussed. Then the suggested solutions to each type
of nodules in recent studies proposed by different authors have been explored with their clinical validations. The
investigations made in this paper are based on the referred articles but being a demanding area of research, the
automated systems of lung cancer need more attention and studies toward this emerging field.
REFERENCES
[1] American Cancer Society, Cancer facts and figures, 2016.
[65] Threshold, For Circularity, Com- 90.41,
classification : pactness, Ellipticity, Sens = 73.55
combination of Eccentricity KNN – Accu =
multilayer per- 91.20,
ceptron (MLP),k- Sens = 81.76.SVM
nearest neighbor – Accu = 90.60,
( KNN) and SVM Sens = 73.44.
classifiers
Aggarwal et al., Threshold, morph- Geometrical & 90 images Cancer Accu = 84 %, Cancer Imaging
2015 [66] ological operation, Statistical features. Imaging Archive Sens= 97.14 % Archive online
Linear discriminate Spec = 53.33 % database
analysis (LDA),
GLCM for comput-
ing features.
Golan et al., 2016 Deep Convolutional Volumetric features NA Sens = 78.9% with LIDC datasets
[67] Neural Network Fp/scan = 20
(CNN), BPN Sens =71.2% with
Fp/scan = 10
Qi et al., 2016 [68] 3D convolutional Shape , Intensity 888 CT scans Sens = 94.4% LIDC datasets,
neural networks and gradient features Manual traced data
(CNNs) and LUNA 16
challenge
Abbreviations: Sen. - Sensitivity = TP/(TP + FN); Spec - specificity = TN/(TN + FP); Accu - Accuracy = (TP + TN)/(TP + TN +
FP + FN); AUC =Area under curve; MSE = mean square error; Fp/scan = no. of false positive per scan. LIDC-IDRI : Lung Imaging
Database Consortium – Image Database Resource Initiative (https://wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI)
(contd...Table 2)
Authors, Year Technique/Method Features Used Database Performance Clinical Validation
classification Accuracy

An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review
[2] S. G. Armato and W. F. Sensakovic, “Automated lung segmentation for thoracic CT,” Acad. Radiol., vol. 11, no. 9, pp. 1011–
1021, 2004.
[3] El-Baz, Ayman, and Jasjit S. Suri, eds. Lung imaging and computer aided diagnosis, CRC Press, 2011.
[4] Mumford, D., Shah, J.: ‘Optimal approximations by piecewise smooth functions and associated variational problems’,
Commun. Pure Appl. Math., 1989, 42, (5),pp. 577–685
[5] Chenyang Xu and Jerry L Prince, “Snakes, shapes, and gradient vector flow,” IEEE Transactions on Image Processing, vol.
7, no. 3, pp. 359–369, 1998.
[6] Laurent D Cohen, “On active contour models and balloons,” CVGIP: Image understanding, vol. 53, no. 2, pp. 211–218,
March 1991.
[7] E. S. Rebouças, R. M. Sarmento and P. P. Rebouças Filho, “3D Adaptive Balloon Active Contour: Method of Segmentation
of Structures in Three Dimensions”
[8] Yuanquan Wang, Lixiong Liu, Hua Zhang, Zuoliang Cao, and Shaopei Lu, “Image Segmentation Using Active Contours
With Normally Biased GVF External Force”. October 2010.
[9] Yuwei Wu, Yuanquan Wang, and YundeJia, “Segmentation of the left ventricle in cardiac cine mri using a shape-constrained
snake model,” Computer Vision and Image Understanding, 2013.
[10] Ahmed Gawish and Paul Fieguth, “external forces for active contours using the undecimated wavelet transform” 2015 IEEE
[11] Tony F. Chan, Member, IEEE, and Luminita A. Vese “Active Contours Without Edges” IEEE TRANSACTIONS ON IMAGE
PROCESSING, VOL. 10, NO. 2, FEBRUARY 2001
[12] Wenchao Cui, Yi Wang*, Tao Lei, Yangyu Fan, Yan Feng, “Local Region Statistics-Based Active Contour Model for Medical
Image Segmentation”, 2013 Seventh International Conference on Image and Graphics
[13] Jiyo S. Athertya, G. Saravana Kumar IEEE Member, “Automatic initialization for segmentation of medical images based on
active contour” 2014 IEEE Conference on Biomedical Engineering and Sciences, 8 - 10 December 2014, Malaysia.
[14] Alaa Khadidos1, 2, Victor Sanchez1 and Chang-Tsun Li1, “Active contours based on weighted gradient vector flow and
balloon forces for medical image segmentation” 978-1-4799-5751-4/14/$31.00 ©2014 IEEE
[15] S. Hu, E. A. Hoffman, and J. M. Reinhardt, “Automatic lung segmentation for accurate quantitation of volumetric X-ray CT
images,” IEEE Transactions on Medical Imaging, vol. 20, no. 6, pp. 490–498, 2001.
[16] Q. Gao, S.Wang, D. Zhao, and J. Liu, “Accurate lung segmentation for X-ray CT images,” in Proceedings of the 3rd
International Conference on Natural Computation (ICNC ’07), vol. 2, pp. 275–279, 2007.
[17] Q.Wei, Y. Hu, G. Gelfand, and J. H.MacGregor, “Segmentation of lung lobes in high-resolution isotropic CT images,” IEEE
Transactions on Biomedical Engineering, vol. 56, no. 5, pp. 1383– 1393, 2009.
[18] X. Ye, X. Lin, J. Dehmeshki, G. Slabaugh, and G. Beddoe, “Shape-based computer-aided detection of lung nodules in
thoracic CT images,” IEEE Transactions on Biomedical Engineering, vol. 56, no. 7, pp. 1810–1820, 2009.
[19] Elham Akbari Baniani and Abdolah Chalechale “Hybrid PSO and Genetic Algorithm for Multilevel Maximum Entropy
Criterion Threshold Selection” International Journal of Hybrid Information Technology Vol.6, No.5 (2013), pp.131-
140
[20] Ming-Huwi Horng, “Multilevel thresholding selection based on the artificial bee colony algorithm for image segmentation”
0957-4174/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.04.180.
[21] Madhubanti Maitra, Amitava Chatterjee, “A hybrid cooperative–comprehensive learning based PSO algorithm for image
segmentation using multilevel thresholding”, 0957-4174/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/
j.eswa.2007.01.002.
[22] SuhuiXu, Xiaodong Mu, Ji Ma, “Discrete Quantum-Behaved Particle Swarm Optimization for 2-D Maximum Entropic
Multilevel Thresholding Image Segmentation” 978-1-4673-7189-6/15/$31.00©2015 IEEE.
[23] Cheol-Hwan Kim, Yun-Jung Lee, “Medical Image Segmentation by Improved 3D Adaptive Thresholding” 978-1-4673-
7116-2/15/$31.00 ©2015 IEEE

Satya Prakash Sahu, Narendra D. Londhe and Shrish Verma
[24] R. Helen, Dr. N. Kamaraj, Dr. K. Selvi, V. Raja Raman, “Segmentation of Pulmonary Parenchyma in CT lung Images based
on 2D Otsu optimized by PSO” 978-1-4244-7926-9/11/$26.00 ©2011 IEEE 536.
[25] Banimelhem, Omar, Moad Mowafi, and Oduy Alzoubi. “Multilevel thresholding image segmentation using memetic
algorithm.” Information and Communication Systems (ICICS), 2015 6th International Conference on. IEEE, 2015.
[26] Shi, Yonghong, et al. “Segmenting lung fields in serial chest radiographs using both population-based and patient-specific
shape statistics.” IEEE transactions on medical imaging 27.4 (2008): 481-494.
[27] Annangi, P., et al. “A region based active contour method for x-ray lung segmentation using prior shape and low level
features.” 2010 IEEE international symposium on biomedical imaging: from nano to macro. IEEE, 2010.
[28] Kockelkorn, Thessa TJP, et al. “Interactive lung segmentation in CT scans with severe abnormalities.” 2010 IEEE International
Symposium on Biomedical Imaging: From Nano to Macro. IEEE, 2010.
[29] Kumar, S. N., and V. Kavitha. “Automatic segmentation of lung lobes and fissures for surgical planning.” Emerging Trends
in Electrical and Computer Technology (ICETECT), 2011 International Conference on. IEEE, 2011.
[30] Sofka, Michal, et al. “Multi-stage learning for robust lung segmentation in challenging CT volumes.” International Conference
on Medical Image Computing and Computer-Assisted Intervention. Springer Berlin Heidelberg, 2011.
[31] Jiantao Pu, David S. Paik, XinMeng, Justus E. Roos and Geoffrey D. Rubin, Shape “Break-and-Repair” Strategy and Its
Application to Automated Medical Image Segmentation IEEE Transactions on visualization and computer graphics, vol. 17,
NO. 1, January 2011.
[32] Sun, Shanhui, Christian Bauer, and Reinhard Beichel. “Automated 3-D segmentation of lungs with lung cancer in CT data
using a novel robust active shape model approach.” IEEE transactions on medical imaging 31.2 (2012): 449-460.
[33] Gill, Gurman, Matthew Toews, and Reinhard R. Beichel. “Robust initialization of active shape models for lung segmentation
in CT scans: A feature-based atlas approach.” Journal of Biomedical Imaging 2014 (2014): 13.
[34] Y. Yim, H. Hong, and Y. G. Shin, “Hybrid lung segmentation in chest CT images for computer-aided diagnosis,” in 7th
International Workshop on Enterprise Networking and Computing in Healthcare Industry, HEALTHCOM2005, pp. 378–
383, kor, June 2005.
[35] A. El-Baz, G. Gimel’farb, R. Falk, T. Holland, and T. Shaffer, “A framework for unsupervised segmentation of lung tissues
from low dose computed tomography images,” in Proceedings of the British Machine Vision Conference, pp. 855–865,
University of Leeds, Leeds, UK, September 2008.
[36] P. Hua, Q. Song, M. Sonka, E. A. Hoffman, and J. M. Reinhardt, “Segmentation of pathological and diseased lung tissue in
CT images using a graph-search algorithm,” in Proceedings of the International Symposium on Biomedical Imaging
(ISBI ’11), pp. 2072–2075, 2011.
[37] De Nunzio, G., Tommasi, E., Agrusti, A., Cataldo, R., De Mitri, I., Favetta, M., Maglio, S., Massafra, A., Quarta, M.,
Torsello, M. and Zecca, I., 2011. Automatic lung segmentation in CT images with accurate handling of the hilar region.
Journal of digital imaging, 24(1), pp.11-27
[38] B. Abdollahi, A. Soliman, A. C. Civelek, X. F. Li, G. Gimel’farb, and A. El-Baz, “A novel 3D joint MGRF framework for
precise lung segmentation,” in Proceedings of the MICCAI Workshop on Machine Learning in Medical Imaging, pp. 86–93,
Nice, France, October 2012.
[39] Cortez, P.C. and de Albuquerque, V.H.C., 2013. 3D segmentation and visualization of lung and its structures using CT
images of the thorax. Journal of Biomedical Science and Engineering, 6(11), p.1099.
[40] Keshani, M., Azimifar, Z., Tajeripour, F. and Boostani, R., 2013. Lung nodule segmentation and recognition using SVM
classifier and active contour modeling: A complete intelligent system. Computers in biology and medicine, 43(4), pp.287-
300.
[41] Kuruvilla, J. and Gunavathi, K., 2014. Lung cancer classification using neural networks for CT images. Computer methods
and programs in biomedicine, 113(1), pp.202-209.
[42] Orkisz, M., Hoyos, M.H., Romanello, V.P., Romanello, C.P., Prieto, J.C. and Revol-Muller, C., 2014. Segmentation of the
pulmonary vascular trees in 3D CT images using variational region-growing. IRBM, 35(1), pp.11-19.

An Automated System for the Detection of Lung Cancer in CT data at Early Stages: Review
[43] Zhou, S., Cheng, Y. and Tamura, S., 2014. Automated lung segmentation and smoothing techniques for inclusion of juxtapleural
nodules and pulmonary vessels on chest CT images. Biomedical Signal Processing and Control, 13, pp.62-70.
[44] Shen, S., Bui, A.A., Cong, J. and Hsu, W., 2015. An automated lung segmentation approach using bidirectional chain codes
to improve nodule detection accuracy. Computers in biology and medicine, 57, pp.139-149.
[45] Jinke Wang and HaoyanGuo “Automatic Approach for Lung Segmentation with Juxta-Pleural Nodules from Thoracic CT
Based on Contour Tracing andCorrection” Hindawi Publishing Corporation Computational and Mathematical Methods in
Medicine Volume 2016, Article ID 2962047, 13 pages
[46] 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 Transactions on Medical Imaging, vol. 22, no. 10, pp. 1259–1274, 2003.
[47] C. I. Fetita, F. Prteux, C. Beigelman-Aubry, and P. Grenier, “3D automated lung nodule segmentation in HRCT,” in Proceedings
of the International ConferenceMedical Imaging Computing and Computer-Assisted Intervention (MICCAI ’03), vol. 2878,
pp. 626–634, 2003.
[48] L. Zhang, M. Fang, D. P. Naidich, and C. L. Novak, “Consistent interactive segmentation of pulmonary ground glass nodules
identified in CT studies,” in Medical Imaging: Imaging Processing, Proceedings of SPIE, pp. 1709–1719, February 2004.
[49] A. P. Reeves, A. B. Chan, D. F. Yankelevitz, C. I. Henschke, B. Kressler, and W. J. Kostis, “On measuring the change in size
of pulmonary nodules,” IEEE Transactions on Medical Imaging, vol. 25, no. 4, pp. 435–450, 2006.
[50] K. Okada, V. Ramesh, A. Krishnan, M. Singh, and U. Akdemir,”Robust pulmonary nodule segmentation in CT: improving
performance for juxtapleural cases,” in Proceedings of the International Conference on Medical Imaging Computing and
Computer- Assisted Intervention (MICCAI ’05), vol. 8, pp. 781–789, 2005.
[51] O. Zinoveva, D. Zinovev, S. A. Siena, D. S. Raicu, J. Furst, and S. G. Armato, “A texture-based probabilistic approach for
lung nodule segmentation,” in Proceedings of the International Conference on Image Analysis and Recognition, vol. 2, pp.
21–30, 2011.
[52] W. A. Browder, A. P. Reeves, T. V. Apananosovich, M. D. Cham, D. F. Yankelevitz, and C. I. Henschke, “Automated
volumetric segmentation method for growth consistency of nonsolid pulmonary nodules in high-resolution,” in Medical
Imaging: Computer-Aided Diagnosis, vol. 6514 of Proceedings of SPIE, February 2007.
[53] Y. Tao, L. Lu, M. Dewan et al., “Multi-level ground glass nodule detection and segmentation in CT lung images,” in Proceedings
of the International ConferenceMedical Imaging Computing and Computer-Assisted Intervention (MICCAI ’09), vol. 1, pp.
715–723, 2009.
[54] J. P. Ko, H. Rusinek, E. L. Jacobs et al., “Small pulmonary nodules: volume measurement at chest CT—phantom study,”
Radiology, vol. 228, no. 3, pp. 864–870, 2003.
[55] J. M. Kuhnigk, V. Dicken, L. Bornemann et al., “Morphological segmentation and partial volume analysis for volumetry of
solid pulmonary lesions in thoracic CT scans,” IEEE Transactions on Medical Imaging, vol. 25, no. 4, pp. 417–434, 2006.
[56] Kawata, Yoshiki, et al. “Classification of pulmonary nodules in thin-section CT images based on shape characterization.”
Image Processing, 1997. Proceedings., International Conference on. Vol. 3. IEEE, 1997.
[57] El-Baz, Ayman, Georgy Gimel’farb, Mohamed Abou El-Ghar, and Robert Falk. “Appearance-based diagnostic system for
early assessment of malignant lung nodules.” In 2012 19th IEEE International Conference on Image Processing, pp. 533-
536. IEEE, 2012.
[58] Han, Fangfang, Huafeng Wang, Guopeng Zhang, Hao Han, Bowen Song, Lihong Li, William Moore, Hongbing Lu, Hong
Zhao, and Zhengrong Liang. “Texture feature analysis for computer-aided diagnosis on pulmonary nodules.” Journal of
digital imaging 28, no. 1 (2015): 99-115.
[59] Dhara, Ashis Kumar, Sudipta Mukhopadhyay, Pramit Saha, Mandeep Garg, and Niranjan Khandelwal. “Differential geometry-
based techniques for characterization of boundary roughness of pulmonary nodules in CT images.” International journal of
computer assisted radiology and surgery 11, no. 3 (2016): 337-349.
[60] Antonelli, M., Cococcioni, M., Lazzerini, B. and Marcelloni, F., 2011. Computer-aided detection of lung nodules based on
decision fusion techniques. Pattern Analysis and Applications, 14(3), pp.295-310

Satya Prakash Sahu, Narendra D. Londhe and Shrish Verma
[61] de Carvalho Filho, Antonio Oseas, Wener Borges de Sampaio, Aristófanes Corrêa Silva, Anselmo Cardoso de Paiva, Rodolfo
Acatauassú Nunes, and Marcelo Gattass. “Automatic detection of solitary lung nodules using quality threshold clustering,
genetic algorithm and diversity index.” Artificial intelligence in medicine 60, no. 3 (2014): 165-177.
[62] Choi, Wook-Jin, and Tae-Sun Choi. “Automated pulmonary nodule detection based on three-dimensional shape-based feature
descriptor.” Computer methods and programs in biomedicine 113, no. 1 (2014): 37-54.
[63] Alilou, Mehdi, Vassili Kovalev, Eduard Snezhko, and Vahid Taimouri. “A comprehensive framework for automatic detection
of pulmonary nodules in lung CT images.” Image Analysis & Stereology 33, no. 1 (2014): 13-27.
[64] Taþcý, Erdal, and Aybars Uður. “Shape and texture based novel features for automated juxtapleural nodule detection in lung
cts.” Journal of medical systems 39, no. 5 (2015): 1-13.
[65] Farahani, Farzad Vasheghani, Abbas Ahmadi, and MH Fazel Zarandi. “Lung nodule diagnosis from CT images based on
ensemble learning.” In Computational Intelligence in Bioinformatics and Computational Biology (CIBCB), 2015 IEEE
Conference on, pp. 1-7. IEEE, 2015.
[66] Aggarwal, Taruna, Asna Furqan, and Kunal Kalra. “Feature extraction and LDA based classification of lung nodules in chest
CT scan images.” In Advances in Computing, Communications and Informatics (ICACCI), 2015 International Conference
on, pp. 1189-1193. IEEE, 2015.
[67] Golan, Rotem, Christian Jacob, and Jörg Denzinger. “Lung nodule detection in CT images using deep convolutional neural
networks.” In Neural Networks (IJCNN), 2016 International Joint Conference on, pp. 243-250. IEEE, 2016.
[68] Dou, Qi, Hao Chen, Lequan Yu, Jing Qin, and Pheng Ann Heng. “Multi-level Contextual 3D CNNs for False Positive
Reduction in Pulmonary Nodule Detection.” DOI 10.1109/TBME.2016.2613502, IEEETransactions on Biomedical
Engineering.