See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/344479889

Lung Cancer Detection System Using Image Processing and Machine Learning
Techniques

Article in International Journal of Advanced Trends in Computer Science and Engineering · August 2020
DOI: 10.30534/ijatcse/2020/260942020

CITATIONS READS
27 8,694

1 author:

Vikul Pawar
Government College Of Engineering Aurangabad
22 PUBLICATIONS 104 CITATIONS

SEE PROFILE

All content following this page was uploaded by Vikul Pawar on 13 February 2021.

The user has requested enhancement of the downloaded file.

 ISSN 2278-3091
Volume
Vikul J. Pawar et al., International Journal of Advanced Trends9,inNo.4, JulyScience
Computer – August 2020
and Engineering, 9(4), July – August 2020, 5956 – 5963
International Journal of Advanced Trends in Computer Science and Engineering
Available Online at http://www.warse.org/IJATCSE/static/pdf/file/ijatcse260942020.pdf
https://doi.org/10.30534/ijatcse/2020/260942020

Lung Cancer Detection System Using Image Processing and

Machine Learning Techniques
Vikul J. Pawar1, Kailash D. Kharat2, Suraj R. Pardeshi3, Prashant D. Pathak4
1
University College of Engineering, Osmania University, Hyderabad, India, [email protected]
2
Government Engineering College, Aurangabad, Maharashtra, India, [email protected]
3
Government Engineering College, Aurangabad, Maharashtra, India, [email protected]
4
Government Engineering College, Aurangabad, Maharashtra, India, [email protected]

 ABSTRACT the stage of disease, health of patient, and some other factors.
In recent time the Lungs Cancer diseases are increases The rate of survival is only 14% of patient for five years. Lung
widely, the human body made up of with diverse fundamental Cancer develops in respiratory epithelium of bronchial tree of
organs, one of those is Lung, The Lungs are of two section lungs. It is very rare to detect the lung cancer before the age 45
right lung and left lung and function of lungs are to trade of years but generally Lung Cancer may be detected in the age 55
gas which we called as breathing or respiration. The present to 70 [1].
modern lifestyle, ecological contamination is colossally In the direction of to prevent the Lung Cancer disease and
expanding the Human Lungs issue. What's more, for analyzing the early stage of this disease required powerful
correspondingly there are various Image processing technology to assist the doctors is mostly desirable, in
techniques are concocting the tremendous answer for the particular Image Processing, Machine Learning and
Medical field to identify and analyze the Lungs illnesses. This
Artificial Intelligence technique can process the medical field
article is contemplating and exploring few of those Image
data with the aid of engineering solution for the purpose of
Processing techniques, The Computer Aided Diagnosis
(CAD) frameworks requires the preprocessing and highlight detection and diagnosing the Lung Disease. It is needed to
extraction from X-rays, Computer Tomography (CT) scan to preprocess and trained the medical field data such X-Rays,
analyze the Lungs disorders in human body, this paper is Computed Tomography (CT) scan images by applying
bringing the emerging image processing and machine various Neural Network, Machine Learning techniques to the
learning techniques for implementing purpose, and by input dataset [2].
comparing the diverse classification techniques will assist to There are varied Lung Disease Diagnosis (LDD)
improve the accuracy in lung cancer detection system using prototypes are developed by the researchers to improve the
robust segmentation and classification techniques. disease detection techniques in early stage of lung cancer,
which will help to the practitioner or doctors. The
Key words: Computer Aided Diagnosis (CAD), X-rays, Computed-Tomography (CT) Scan images are the mainly
Computer Tomography (CT), Lung Disease Diagnosis suitable for the invention of pulmonary nodule in Lung
(LDD), Image Processing Technique, Lung Cancer, Cancer [3] [4].
Classification. The small pulmonary nodule can be simply detected and
eventually early disorders in nodule size and number can be
1. INTRODUCTION detected through three dimensional Computed Tomography
In the past few years the Lung Cancer became the major (CT) image [5][6]. The foremost objective of this study is to
health diseases in human body. It is quite difficult to diagnose summarize a variety of review articles on Lung Cancer
Lung Cancer in early stage of it, which may leads to increase Diagnosis and proposing the robust segmentation and
the risk factor of survival of patients. Correspondingly the classification technique.
treatment on Lung cancer depends upon the how early this
disease can be diagnose so that treatment can control on 2. RELATED WORK
increasing (in stage) and spreading of Lung Cancer in other The literature review on Lung Disease Diagnosis (LDD) is
part of body. It is quite possible to control Lung Cancer going to study the various article related to Lung Cancer
disease by giving proper treatment, there are various Disease which applies image processing techniques on
treatments are available in the field of Medical Science such Computed Tomography (CT) and X-rays images.
as Surgery, Chemotherapy and radiography as it is depend on Xiaodan Chen S. Feng, and D. Pan. (2015) [7], Author of
this paper has applied the adaptive threshold algorithm,
mathematical morphology and Watershed algorithm to

5956
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

overcome the issues faced while diagnosing lung disease successfully detect the cancer nodule but still there is scope of
through CAD system, The researchers of this article have improvement in accuracy, this model has used simple
applied the few steps for image segmentation method; in first segmentation technique there is extent to have some trained
step they have enhanced the quality of image by applying the learning machine to progress the classification of the
Gaussian filter and gradient enhancement method to reduces cancerous nodules.
the noise from original Computed Tomography image(CT) P. B. Sangamithraa and Govindaraju, S (2016) [10], in
then in second step the images are segmented with OTSU this proposed model initially CT images are pre-processed for
method thereafter in third step they have removed the main removal of noise, in next stage the fuzzy k-means algorithm
bronchus and trachea from CT lung images and then at last are used to segment the tumor from lungs, then further result
segmented with improved watershed transform, and testing is of segmentation is improved by using K-means approach after
performed using series of Computed Tomography (CT) that in next step some features of lungs are extracted from CT
Lungs images to segment the area of lung parenchyma, at the images such as entropy, correlation, homogeneity, SSIM and
result of this researcher have experimentally shown that how PSNR, the feature extraction used the statistics methods
lung parenchyma is segmented successfully. known as gray scale co-occurrence matrix(GLCM) and in
Manikandan T (2017) [8], Author is stated that the CAD final step categorization is performed by using supervised NN
(Comp. Assisted Diagnosis) model can be helpful to solve the neural network like Back Propagation BPNN for
lung cancer recognition in its early stage using CT images, in identification of cancerous lungs. This model achieved the
his paper author directed that the major four steps are there to accuracy of 90.7%, still the accuracy can be improve by
detect the nodule in lung cancer patient, author suggested that developing improvise classification technique such as
there are certain challenges will be faced by medical Support Vector Machine.
professionals while diagnosing the cancer nodule is complex Jin, X., Zhang, Y., & Jin, Q. [11], to detect Lung Cancer
from normal CT images, so that these four stages will be the proposed model used C Neural Networks (convolution
impactful to detect the lung cancer, in first step need to NN) as classification approach in the Computer Aided
remove the noise by applying various filter like mean filter, Diagnostic (CAD) system. And this model has reached and
median filter, Gaussian filter, Weiner filter, Min filter, Max achieves accuracy of 84.6%, sensitivity 82.5% and
filter, Gabor filter, in second step there is segmentation of specificities 86.7%. Still the accuracy is unsatisfactory; the
suspected nodule of lung cancer using different segmentation benefit of proposed system is that it use the circular filter in
technique such as multiple thresholding, optical focused regions while extracting that helps to reduce the
thresholding, global thresholding, active contour method, overall cost of the detection and training stage.
morphological method, watershed segmentation, shape-based Zakaria Suliman Zubi and et al.[12], proposed the few
method, template matching, the third step is to extracting the methods of the data mining. The lung cancer database of
features (2-D features) such shape-size features, geometric patients consists of medical images Human upper body X-ray
features, gray scale feature, gradient features also some that classifies in assorted three categories; which are
statistical features can be extracted from lungs to detect the malignant, normal and benign. The healthy patients are non
nodules are cancerous or noncancerous, and final step is of nodules which are categorized as normal patients; the
classification of extracted features from lungs belongs to lung_nodules may be initial benign stage or a normal lung
malignant or benign, in this steps various classifier can be without cancer, and malignant are Cancerous patients. The
applied such as support vector machine, Linear Discriminate CAD system works in different steps, the pattern recognition
Analysis, Generic Algorithm, Rule Based Classifier, is the formation of a feature extraction process and the
Artificial Neural Network (ANN) were implemented classification process. Proposed system used X-ray chest films
successfully, as described by author in his paper still there is which obtained the less accuracy compare to CT images,
lot of improvement need to be done in Sensitivity, Specificity study states that the future work implementation extends to
and Accuracy in existing diagnosis system. use the CT images for superior diagnosis of Lung Cancer
T Aggrawal et al. (2015) [9], In their research it is proposed detection.
that, a system for the classification of cancer nodule and also a Xinyan Li S.F., Daru Pan. (2016) [13], has developed the
normal lung anatomy, author of this model used the most enhanced method for segmentation approach of the lungs
valuable thresholding values and gray scale characteristics to CT-Image, the researcher of this article have combines the
process the segmentation approach of the lung nodule then kernel graphs cut algorithm and mathematical solution and
the geometric features are extracted from nodule and also this algorithm is comparatively studied with K-means
extracted geometric features are combined with LDA algorithm and OTSU’s maximum between-cluster variance
classification method for finding the difference between algorithm, the researchers of this article have applied the
normal structure of the lung and cancer nodules. The kernel graph cuts method for image segmentation from
projected system achieved 84% of accuracy, 97.14% Computed Tomography (CT) Lungs images and they have
sensitivity, 53.33% specificities, however the proposed model proposed their method by combing the kernel graphs cut

5957
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

algorithm and mathematical solution and also they have This method used without human interference and method
compared their proposed algorithm with K-means and OTSU seems it is more flexible when it depends on the dose of the
algorithm at last in the experimental results they have shown high radiations also on low radiation dose; the purpose of this
in comparison is that the accuracy of two other methods is method is to apply the automatic identification of shape of
higher. lung with the same it also recovers the unfilled area of lung
Hanan M. et al (2018) [14], In this proposed system with and make the input dataset is ready for CAD system.
the support of comp. assisted diagnosis CAD) system Ahmed, Amer, and F.E.Z. Chadi (2019) [17], In this paper
researchers are targeting to discover the Cancer Nodules from authors has developed the optimizing technique to detect the
Computed-Tomography(CT) images, this system is working pulmonary nodules using deep learning techniques in early
in four stages: in the first stage preprocessing on Computed stage, at the outset they have started to preprocessing of input
Tomography (CT) images is conducted which will help to image through the preprocessing image contrast is adjusted
improve the contrast of image and noise removal from input for low dose scan, then at second step the authors applied a
image dataset in Second step the system performs the transfer learning for the purpose feature extraction. Then in
segmentation of pulmonary nodules and blood vessels by third step genetics algorithm GA is applied to the features
applying double level of thresholding in addition to the help which are extracted by using deep learning approach and also
of morphological operation, in third step feature fusion training subset of data for getting significant pulmonary
technique is to apply for extracting the features from nodules from optimized extracted features. And the final step
segmented image, the feature fusion is made up of four feature is of classification of selected features is performed for
extraction methods which are valued histogram (VH) feature, recognizing the lung cancer using SVM
histograms of oriented gradients (HOG) feature, the statistical Support_Vector_Machine and using ECLAP online lung
feature of first and second order and texture Feature of the image database the experimental result of this prototype
gray scale co occurrence matrix(GLCM), at last in the fourth achieved the detection accuracy 92.5%, Sensitivity is 90%,
step to get the superior accuracy there are three classifier are and specificity is of 95%.
bring into play which are Multilayer’s FF forward NN Suren Makaju et al. (2018) [18], In this paper the
(MffNN), the second classifier is a neural network function researcher has applied the new technique for diagnosing the
with radial basis (RBFNN), the third classifier is used as SVM lung cancer in early stage of it, researcher has used the
Support_Vector_Machine. The quantitative parameter used Median filter and Gaussian filter for preprocessing of
for proposed system for validating the accuracy through Computed Tomography (CT) images rather than Gabor filter
classifications accuracy rates (CAR), Specificity (SF), also then in next stage the images are segmented using watershed
sensitivity (S), around forty CT images were tested for the algorithm, up to this stage authors can identify cancer nodules
purpose of testing and final result achieved by this system is marked. In addition to authors have extracted the feature like
the CAR-99.06%, S-100% and SP-99.2%. But as per finding eccentricity, perimeter and area, pixel_mean_intensity,
of this article there is scope of improvement in identifying Centroid and diameter, thereafter the classifier are used for
benign and malignant tumor. the purpose of classification of cancer nodule using SVM, by
Silva, Carvalho, and Gattass (2004) [15], This paper training the prototype model is able to the classification of
presents how the lung nodule is being diagnose using Gini cancer nodule. For the Implemented purpose authors have
Coefficient and skeletonization methods, the purpose of this used Lung Image Database Consortium (LIDC) [19]. The
method is used to characterize the lung cancer nodules as authors of this paper have increased the accuracy of detection
class malignant or as normal class benign in CT image lung cancer and classify the benign or malignant in lung
dataset, in this article the researcher are initially analyze the cancer. This paper has reached the accuracy of lung cancer
texture by applying Gini coefficient, this method help to detection up to 92% from earlier detection value 86.6%, but
analyze the distribution of nodules in lung with the same there still there this paper shows the scope of improvement in
is another method is used in this paper that is skeletonization increase in accuracy and detection of various stages of lung
which analyze the shape of the nodules, through this authors cancers like stage-I, Stage-II, Stage-III and Stage-IV which
has developed their prototype and using the discriminant really can help the medical field to diagnose the lung cancer
analysis to categories the lung nodule as malignant or benign. and take appropriate decision in treatment to the patient.
The evaluation of the result was based on classification and Geraldo et al. (2014) [20], This paper aims to identify the
ROC curve. COPD and fibrosis lung disease, for the segmentation of lung
Michela Antonelli B. Lazzerini, and F. Marcelloni. (2005) structures researchers has develop the new method is called
[16], The author of this paper has applied image processing the Adaptive Crisp Active Counter Models (ACACM), then
techniques for automatic identification pulmonary on the basis of various attribute of lung images a spatial inter
parenchyma this method combine the thresholding, opening dependence matrix analyzes structural information’s of the
closing morphological operation, border thinning, edge lung pictures at last the classification step classify the lung
recognition, edge reconstructing and also filling the region. disease and health lungs as a result of this model gives 96%

5958
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

accuracy to detect the normal lungs and lungs with COPD X-ray images. The flow of steps to be performs in Image
disease and hereby it is concluded by author is the proposed Processing techniques are represented in figure No 1.
method is feasible assist the diagnosis of disease accordingly.
Zhuoqi and S.H. Idiopathic. (2019) [21], The main purpose The steps to be perform on Computed Tomography or X-ray
of this paper is to study the CNN structure selection and images:
comparison of effort, for detection candidate the utilization of i. Image Acquisitions
Fast_RCNN and Faster_RCNN, to increase the network ii. Image Enhancements
depth use residual learning neural network, for the iii. Image Segmentations
initialization of parameter how to use the transfer learning, iv. Feature Extracting
for optimizing the network training use of v. Classifications of Tumor
network_parameter_iterative_method all these approaches
are applied to analysis, detection lung nodule, and for The Computed Tomography (CT img) and x-ray image is
comparative study. In neural network the transfer learning used as input to Comp. Assisted Diagnosis (CAD) model for
methods produces good result but still there are few other the preprocessing, segmentation, feature extraction,
methods can be elaborated to achieve greater result. This classification and analyze the final result for the judgment of
article created the scope for future work for the improvement lung cancer.
of classification is how effectively can be use the cutting_edge
deep_learning method. i. Image Acquisitions
Allison and A.M.R.W. (2017) [22], In the proposed model The process of reading the input images from some defined
authors has implemented the deep learning technique to source for the further processing, this is initial step of Image
improve the accuracy of diagnosis of lung cancer using CNN processing technique. The image size of 512 x 512 pixels has
network with manifold preprocessing methods [23-25]. The Gaussian noise in input CT image, it is possible to remove the
proposed method is different from others is due to use of use of Gaussian noise spatial filtering by smoothing the image.
two network with different type of inputs and for final
prediction authors have used the voting system. The steps ii. Image Enhancements
under this model by using Gaussian filter firstly performs the This Image Enhancements process is used for making image
preprocessing on computed tomography images that CT more comprehensible and undoubtedly detectable for
images will go thru the randomize testing then these image forthcoming analysis of digital image. To get the more
will be used as training dataset for CNN network, then in outstanding result in segmentation it is required to enhance
second stage smoothed and unsmooth images are combined, the image. The essential part defines the Laplacian_operator
thereafter in third stage using convolution filter the feature input image to detect the image in equation (1). Output of this
will go to the maximum pooling layer then the output of third will get the Edge enhanced image [27].
stage that is features will be the input to the softmax also to
the fully connected layer. In this way every CNN network will -1 -1 -1
be created and at the final stage the voting system produces -1 8 -1
the output. The accuracy achieved by this model was 97.5% -1 -1 -1 --------------- (1)
and the false positive rate is under 10%, the authors are
expecting to increase the accuracy which needs to modify in At the second stage, scaled image of earlier stage image is
voting system. converted to gray scale image as per in equation (2).

3. IMAGE PROCESSING ARCHITECTURE EI = I + k*LI --------------- (2)

The objectives of this annotations is to analyze and do the EI : is edge enhanced image,
study of various model proposed by esteemed authors in their I : is original lung image,
respective article, this paper enlighten the issues on Lung LI : images acquired by convolving
Disease Diagnosis (LDD) using Computed Tomography (CT) I : with Laplacian operator,
and X-ray images. The Image Processing Technique is the k : is a constant.
most popular because of its versatility for developing superior
model and finding out the abnormality in different Computed iii. Image Segmentations
Tomography or X-ray images [26]. The image processing The image segmentations is the course of action which
technique is performing the an assortment of steps like IA separates the images in to the different section that can
acquisition of Image, IE Enhancing the image quality, IS facilitate to extract the interested information from input
segmentation of Images, FE extracting features, image. The Lung Segmentation method based on the
Classification and Conclusion on Computed Tomography or thresholding technique initially introduced by Hu et al

5959
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

(2001)[28]. Otsu’s et al. (1979) [29], has indomitable the extracted. There are distinct feature extractions methods are
global threshold value as T, using T the image is segmented used by researchers few of those methods are Histogram of
and two group of pixels is formed by this method [30]. oriented gradients features (HOGs) [31], the first and second
orders statistical features [31], texture features of the gray
G1=pixels_intensity_values are greater than and equal to scale co occurrence matrix [32] on the basis of wavelets
T, coefficient, and value histograms (VH) [32].

G2=pixel values are less than T. v. Classifications of Tumor

At this stage of classification the classifiers are labeling the
Then the threshold values are computed for the pixel images in several consortiums to analyze the final result on
regions G1 & G2 using the average intensity value µ1 and µ2. digital image. The widely used classifiers for predictive
analysis of Lung cancer detection are SVM
T=½(µ1+µ2) ---------------------------- (3) Support_Vector_Machine [33], K-NN K-nearest neighbors
The process of computing value of T can be repeated until. [34], Decision tree and Artificial Neural Networks
Once the convergence is reached the threshold value will be (ANN)[35]. This entire classification algorithm’s
set as Tc, then to indicate the body-pixel any intensity_pixel performance is measures on the basis of parameter of
higher than Tc will be set as 0 and for non-body pixel the sensitivity, accuracy and specificity of proposed system.
other pixel values set as to the 1. The output of this step will At the end the comp. assisted diagnosis model (CAD) is to
come out as Threshold image. Simultaneously the further compare the obtained result from classification step for Lung
stages in segmentation are background elimination. Disease Diagnosis (LDD) and produces the final result. The
figure 1 and figure 2 are the (courtesy of LIDC-IDRI dataset)
CT /X-Ray
Images

Image Acquisition

Smoothing Enhancement Segmentation

Feature Extraction
Figure 2: Cancer Patients Lungs CT Image (LIDC-IDRI dataset)
[19]

Database Classification

Benign Mild Malignant

Figure 1. The steps of Image Processing Technique for Lung

Disease Diagnosis (LDD) using Computed-Tomography (CT) and
X-ray images

iv. Feature Extracting

Features are nothing but the some measured values or
attributes that can be informative for ranking the attribute for
predictive analysis. On the basis pixel based information the
Figure 3: Cancer Patients Lungs CT Image with highlighting spot of
section of attention from segmented lungs image can be
cancer nodule (LIDC-IDRI dataset)

5960
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

of low dose Lung CT images of cancer patient, normally it is implement and speedy classifier with good interpreter,
challenging for medical professionals to detect the cancer moreover the complexity of these classifier goes increases
nodule and come to the final decision, the red spot with increase in dataset size.
highlighted in figure are the cancer nodules, however it is
complex to conclude the exact nodules, the proposed system Table 2: Comparative study of established classification methods:
performs the segmentation and classifies the cancer nodule
using R-CNN robust classification techniques. Methods/
Advantages Disadvantages
Classifiers
4. DISCUSSION ON METHODS AND ALGORITHMS
Support a) Highly applicable a) It required the image
Vector with less sample intensity value.
The research area of image processing needs to apply the Machine of data. b) This is
various methods at the different stage of processing, to make (SVM)[36] b) Healthy functional comparatively slow
the images unambiguous and significant to the techniques, [37] for the data with in meeting the
the low dose Computer Tomography (CT) images may have high dimension. expectation for
larger dataset.
in indistinguishable samples. The ambiguity in image can be
Artificial a) It imitate in a) Only applicable for
resolved using preprocessing technique as discussed in Neural periodic manner. larger data sample.
previous chapter of this article. Network b) It is adequately b) Difficulty to develop
(ANN)[37] grip the larger the prototype
4.1. Experimental study data sample. formation.
The diverse techniques are used by the researchers for Decision Tree a) Easy to implement. a) It has voracious
(DT)[37] b) The ability to characteristics.
producing high-quality result in detection of lung cancer,
integrate the b) Sometimes the
correspondingly as SVM Support_Vector_Machine [34], multiple classifiers create
K-NN K-nearest neighbors [34], Decision tree and Artificial interpreters in an more complexity
Neural Networks (ANN), RBF-ANN[35]. The result of these easy and step by which leads
classification techniques was adequately statistically step manner. challenges to
compared and presented by Hanan et al. (2018)[14]. In table represent the
result.
1 it is represented that the how result is modified prior and
K-Nearest a) This is faster due a) It is not compatible
subsequent use of Genetic Algorithm, from this it has been Neighbor to no training for large dataset.
understand that the SVM and ANN has similar values of (KNN)[38] time. b) KNN is less
Accuracy and Specificity Hanan et al. (2018). b) KNN is easy to functional with
implement. noisy data.
Table 1: Shows the number of features extracted prior and later Convolution a) Fully automatic, a) CNN is invariant to
using GA Algorithm, achieved Accuracy, Sensitivity and Neural manual rotation and scale.
Specificity[14]. Network intervention is b) Prone to overfitting
(CNN)[39] not required. due to complexity
Extracted Good classifier at of model
Size of
Classification Hybrid Sensiti- Specific structure.
Technique Feature
Selected Accuracy
vity ity
feature detection
Feature
Size
b) High accuracy in
SVM 182 153 96.6% 100% 94.2%
image processing
RBF-ANN 182 115 95.2% 100% 97.4% classification.
ANN 182 169 94.6% 100% 95.2% Naïve Bayes a) Easy to Implement a) Assumes
[40] and robust to independence of
irrelevant features.
attributes. b) Performance
4.2. Comparative Study of Classification Methods b) Tiny amount of degraded when
There are several well known and successfully dataset is features are highly
implemented methods/classifiers are discussed in this section, required to run correlated.
this study gives more prominence on functioning of the this classier
classifier which assists the researchers in conclusive step of
the prototype development. The outcome of comparative
study between SVM Support_Vector_Machine [36][37] and 5. CONCLUSION
Artificial Neural Networks (ANN) [37] the both classifiers
are effective in functional whereas these classifier works The analysis on CT image segmentation is observes that the
better at the variance in dataset size is take place. The K-NN effective use of different techniques such as thresholding,
K-nearest neighbors [38], Decision tree [37] are easy to classification techniques such ANN, SVM and etc. achieves

5961
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

the objectives of researchers are up to the mark, but it is 4. Howe, M.A. and B.H. Gross. CT img evaluations of the
observed that still there is scope of improvement in accuracy, equivocal pulmonary nodules, Comput Radiol, 1987.
in detecting the lung cancer disease. In several studies it is 11(2): p. 61-7.
inspected that the lung cancer may detected in last stage of it, 5. Yankelevit, D.F, et al., A Small pulmonar nodules:
which may cause of increase in risk of death of patient, as per volumetrically determined growth rate based on CT
medical term it is very much clear that the lung cancer can be img evaluations, Radiology, 2000. 217(1): p. 251-6.
treatable if it can be detected in early stage where as the 6. Ko, J.P. and M. Betke. Chests CT img: An automated
several study says that sometime lung cancer doesn’t give any nodules detections and assessments of change over
symptoms in early stage of it, and similarly noisy CT image time preliminary experience, Radiology, 2001.
doesn’t give appropriate result in lung disease using computer 7. Chen, S. Feng, and D. Pan. An improved approach of
lungs image segmentations based on watershed
diagnosis system (CAD). There is need of such CAD system
algorithms, 7th International Conference on Internet
to work on robustness of preprocessing techniques,
Multimedia Computing & Service - ICIMCS '15. 2015.
segmentation technique, noise removal methods, and
Zhangjiajie, Hunan, China: ACM New York, NY, USA
accurate diagnosing classifier to achieve the objectives of ©2015.
Medical Imaging field by applying Image Processing 8. Manikandan T. Challenges in the lungs cancers
technique. The proposed system overcome the challenges of detections using computer aided diagnosis system – a
preceding methods which are used in detection system and key for survival of patients, Arch Gen Intern Med 2017
exploit the robust noise filtering methods using Autoencoder Vol-1, Issue-2.
system, Segmentation and Classification techniques such as 9. Aggarwal, T., Furqan, A., & Kalra, K. Features
OTSU algorithm for segmentation technique and extractions and LDA based classifications of lungs
classification techniques are Decision tree and CNN. On the nodules in chest CT scan image, 2015 International
basis of truthful study the comparative statement on Conference On Advances In Computing,
classification techniques are expressed in previous section, Communications And Informatics (ICACCI).
this paper has given brief discussion over the how Lung https://doi.org/10.1109/ICACCI.2015.7275773
Disease Diagnosis (LDD) research is progressing to reach the 10. Sangamithraa, P., & Govindaraju, S. Lungs tumour
fulfillment of medical field using Robust R-CAD system, and detection and classifications using K-Mean
at the same time how challenges can be overcome in terms of clustering, International Conference on Wireless
future work. Communication, Signal Processings And Networking
(Wispnet), 2016.
11. Jin, Zhang, Y., & Jin. Pulmonary Nodules Detections
ACKNOWLEDGEMENT Based on CT Image Using Convolutional Neural
Networks, 9th International Symposium on
We authors of this research paper are grateful to Dr. P V Computational Intelligence And Design, 2016.
Sudha, Professor & Head in CSE department, University 12. Zakariya Suleman Zubi and Rema Saad. Improves the
Engineering college Osmania University, Hyderabad, India. Program of Lungs Cancers using data mining
Dr. K. Shyamala, Professor & Dean in CSE department, Technique, Journal of Software Engineering and
University Engineering college Osmania University, Applications, 2014.
Hyderabad, India, for their continuous motivation, help and 13. Xinyan Li, S.F., Daru Pan. Enhanced lungs
always being there to support segmentations in chests CT images based on kernel
graphs cut, International Conference on Internet
Multimedia Computing and Service. 2016. Xi'an, China:
REFERENCES
ACM New York, NY, USA ©2016.
14. Amer, H.M., et al., A Computer Aided Early Detection
1. Witschi, H. A short history of lungs cancer, Toxicol model of Pulmonary Nodules in CT Scan Image, 7th
Sci, 2001. 64(1): p. 4-6. International Conference on Software and Information
2. Furat, et al., Machine-Learning Techniques for the Engineering - ICSIE '18. 2018, ACM: Cairo, Egypt. p.
Segmentations of Tomographic Images Data of 81-86.
Functional Materials, Frontiers in Materials, 2019. 15. Silva, A.C., P.C.P. Carvalho, and M. Gattass. Diagnosis
6(145). of lungs nodules using Gini coefficients and
https://doi.org/10.3389/fmats.2019.00145 skeletonization in computerized tomography image,
3. E.R.M., The roles of computed tomography in the Proceedings of the 2004 ACM symposium on Applied
diagnosis and management of lungs cancer: an computing - SAC '04. 2004: Nicosia, Cyprus.
overview, The American College of Chest Physicians. https://doi.org/10.1145/967900.967954
Published by Elsevier Inc., 1986. 89(4): p. 5. 16. Antonelli, M., B. Lazzerini, and F. Marcelloni,
Segmentations and reconstructions of the lungs

5962
Vikul J. Pawar et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(4), July – August 2020, 5956 – 5963

volume in CT image, Proceeding of the 2005 ACM https://doi.org/10.1109/TSMC.1979.4310076

symposium on Applied computing - SAC '05. 2005. 30. Pawar, Vikul., An Improved Binarization Techniques
17. Elnakib, A., H.M. Amer, and F.E.Z. Abou-Chadi, For Degraded Documents Images Using Local
Computer aided detection systems for early Thresholding Methods, IJPARET International Journal
cancerous pulmonary nodule by optimizing deep of Pure & Applied Research In Engg. & tech., 2014. 2(8):
learning feature, 8th International Conference on p. 71-82.
Software and Information Engineering - ICSIE '19. 31. Madero Orozco, H., et al., An Automated systems for
2019, ACM Transaction: Cairo, Egypt. lungs nodule classifications based on wavelet feature
18. Makaju, S., et al., Lungs Cancer Detections using CT descriptors and support-vector-machines, Biomed
Scan Image, Procedia Computer Science, 2018. 125: p. Engg Online, 2015. 14(9): p. 9.
107-114. 32. Liu, X., et al., Recognizing the common CT image
19. Armato, McLennan, Gray, Michael, Charles, Reeves, signs of lungs disease through the new features
Anthony P., Clarke, Laurenc, Data from LIDC-IDRI. selection methods based on Fishers criterion and
The Cancers Imaging. 2015. genetic optimizations, IEEE Biomed Health Inform,
20. Ramalho, G.L.B., et al., Lungs disease detections using 2015. 19(2): p. 635-47.
feature extractions and extreme learning machine, 33. Kollabhanu Prakash, Lakshmi Kalyani,Naga Pawan,
Revista Brasileira de Engenharia Biomédica, 2014. “Analysis of Mammography for Identifying Cancer
https://doi.org/10.1590/rbeb.2014.019 Cells Using Convolution Neural network”, IJATCSE
21. Zhuoqi Sun, S.H. Idiopathic Interstitial Pneumonias March –April 2020.
Medical Image Detections Using Deep Learning https://doi.org/10.30534/ijatcse/2020/44922020
Technique: A Survey, ACM Southeast Conference 34. Gray, Givens, Keller, J.M., M.R. A fuzzy K-NN
(ACMSE 2019), April 18-20, 2019, : ACM, New York, algorithm, Transaction IEEE on System, Man, and
NY, USA. Cybernetics, 1985. SMC-15(4): p. 580-585.
22. Zhou, A.M.R.W. Deep Learning for Categorizations 35. Murty, Rao, M.V., N.V., Early Lung Cancers
of Lungs Cancer CT Image, IEEE/ACM International Detections by using Radial Basis Functions Neural
conference on Connected Health: Applications, Systems Network, 2(8), 2015.
and Engineering Technologies (CHASE). 2017. 36. CD., Y.-W., Machine learning with R cookbook. Packt
Philadelphia, PA, USA: IEEE. Publishing Ltd. 2015.
23. R. Fakoor, A. Nazi, and M. Huber. Using deep learning 37. Harrington, P., Machine Learning in Action, ACM
to enhance the cancers diagnosis and classifications, Trans. 2012: ACM.
International Conference on Machine Learning, 2013. 38. M. Clark., Intro. To Machine Learning: with an
2013, MLR:W&CP: Atlanta, Georgia, USA. application using R, 2013.
24. R. Biswas and S. Roy, “Content Based CT Image Sign 39. Jianxin Wu. Convolutional neural networks, 11 Feb
retrival Using Fast Discrete Curvelet Transform and 2020.
Deep Learning” , IJATCSE May-June 2019. 40. Irina and Rish. Empirical study of the Naïve Bayesian
25. F. Ciompi, K.C., S. J. van Riel, Setio, P. K. Gerke, C. classifiers, International journals CAI Workshop on
Jacobs, E. T. Scholten, C. Schaefer-Prokop, M. M. Wille, Empirical Methods in artificial intelligence, 2001.
A. Marchiano. Towards automatic pulmonary nodules
management in lungs cancer screening with deep
learning, 2016.
26. Chaudhary, A. and S.S. Singh. Lungs Cancer
Detections on CT Images by Using Image Processing,
International Conference on Computing Sciences. 2012,
IEEE: India. p. 142-146.
27. Elizabeth, D.S., et al., A novel segmentations approach
for improving diagnostics accuracy of CAD system
for detecting lungs cancer from chest computed
tomography image, Journal of Data and Information
Quality, 2012.
28. Hu, S., E.A. Hoffman, and J.M. Reinhardt. Automatic
lungs segmentations for accurate quantization of
volumetric x-ray CT image, IEEE Trans Med Imaging,
2001.
29. Otsu, N., A Threshold Selection Methods from Gray
scale Histograms, IEEE Transactions on Systems, Man,
and Cybernetics, 1979.

5963

View publication stats