IRJET- Lung Cancer Nodules Classification and Detection using SVM and CNN Classifiers

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 07 | July 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 23
Lung Cancer Nodules Classification and Detection Using SVM and CNN
Classifiers
Yashaswini S.L1, K.V Prasad2
1M.Tech student, Dept of ECE, Bangalore Institute of Technology
2Professor & Head, Dept of ECE, Bangalore Institute of Technology
---------------------------------------------------------------------------***------------------------------------------------------------------------------
Abstract - Cancer isaquite commonanddangerousdisease.
The various methods of cancer exist in the worldwide. Lung
cancer is the most typical variety of cancer. The beginning of
treatment is started by diagnosing CT scan. The risk of death
can be minimized by detecting the cancer very early. The
cancer is diagnosed by computed tomography machine to
process further. In this paper, the lung nodules are
differentiated using the input CT images. The lung cancer
nodules are classified using support vector machine classifier
and the proposed method convolutional neural network
classifier. Training and predictions using those classifiers are
done. The Nodules which are grown in the lung cancer are
tested as normal and tumor image. The testing of the CT
images are done using SVM and CNN classifier. Deep learning
is always given prominent place for the classification process
in present years. Especially this type of learning is used in the
execution of tensor Flow and convolutional neural network
method using different deep learning libraries.
Key Words: Lung cancer, deep learning, biomedical image
classification, confusion matrix, microdicom.
1. INTRODUCTION
Lung cancer is recognized as the main reason behind the
death caused due to cancer in the worldwide. And it is not
easy to identify the cancer in its early stages since the
symptoms doesn’t emerge in the initial stages. It causes the
mortality rate considered to be the highest among all other
methods of cancer. The number of humans dies because of
the dangerous lung cancerthanothermethodsofcancersuch
as breast, colon, and prostate cancers. Thereexist enormous
evidence indicating that the early detection of lung
cancer will minimizemortalityrate.Biomedicalclassification
is growing day by day with respect to image.Inthisfielddeep
Learning plays important role. The field of medical image
classification has been attracting interest for several years.
There are various strategies used to detect diseases. Disease
detection is frequently performed by observant at
tomography images. Early diagnosis must be done to detect
the disease that isleading to death. One among thetoolsused
to diagnose the disease is computerized tomography. Lung
cancer takes a lot of victims than breast cancer, colon cancer
and prostate cancer together. This can be a result of
asymptomatic development of this cancer. The Chest
computed tomography images are challenging in diagnostic
imaging modality for the detection of nodules in lung cancer.
Biomedical image classification includes the analysis of
image, enhancement of image and display of images via CT
scans, ultrasound, MRI.Nodules within therespiratoryorgan
i.e. lung are classified as cancerous and non-cancerous.
Malignant patches indicate that the affected person is
cancerous, whereas benign patches indicate an affected
person as a non- cancerous patient. This can be done using
various classifiers.
2. EXISTING SYSTEM
Support Vector Machines is a method of machine learning
approach taken for classifying the system. It examines and
identifies the classes using the data. It is broadly used in
medical field for diagnosing the disease. A support-vector
machine builds a hyper plane in a very high or infinite-
dimensional area, which can be utilized
for classification, regression, or totally different operation
like outliers detection.
Fig -1: The SVM classifier representation.
Based on a good separation is obtained by the hyperplanein
the SVM. After classification if the gap is large to the nearest
training-data pictures of any class referred as functional
margin, considering that in generally the larger the margin,
the lesser the generalization error of the classifier. Fig-1
shows the support vector machine classifier that constructs
a maximum margin decision hyper plane to separate two
different categories. SupportVectorMachineisalinearmodel
applied for the classification and regression issues.

Fig -2: Training and prediction using SVM
SVM algorithm finds the points closest to the line from both.
The classes of these points are referred as support vectors.
The mixed data of tumor nodules and normal nodules are
provided as input In SVM algorithm the input images given
are trained and the results are predicted, tuning the various
parameters. Fig-2 shows the training and prediction using
SVM. Input images undergofeatureextraction.Atthetraining
the various SVM parameters are tuned, and then the
predictions are made using the hyper plane of SVM.
Fig -3: Testing phase of SVM
Fig-2 shows the training and prediction using SVM. Input
images undergofeatureextraction.Atthetrainingthevarious
support vector parametersaretunedthenthepredictionsare
made using the hyper plane of SVM. In the testing phase the
nodules in the lung cancer are classified as normal or tumor
nodules. Fig 3 shows testing phase of SVM. Initially the input
images are pre-processed. Later SVM operation takes place.
The cancer nodules undergo testing processes. The CT
scanned image undergoes median filtering and estimates
whether the nodules are malignant or benign. Then the
output will be shown as normal image or tumor image.
3. PROPOSED METHODOLOGY
Convolutional neural networks encompassofmultiplelayers
in its structures. CNN could be feed forward and extremely
tremendous approach especially in detection. Network
structure is built easy; has less training parameters. A
convolution neural network have multiple layers within the
neural network, that consists of one or a lot of convolution
layers and so succeeded by one or more fully connected
layers as in a standard multiple layers in neural network.
Convolution neural network architecture is typically
employed collaboration with the convolution layer and pool
layer. The pooling layerisseenbetweenconvolutionlayers.It
confuses the features of the particular position. Since not all
the location features are not important, it just needs other
features and the position. The pooling layer operation
consists of max pooling and means pooling. Mean pooling
calculates the average neighbourhood inside the feature
points, and max pooling calculates the neighbourhoodinside
a maximum of feature points.
Fig -4: Structure of CNN model
A CNN uses the learned features with input and make use of
2D convolutional layers. This implies that this type of
network is best for processing2Dimages.Comparedtoother
methods of image classification, the network uses very little
pre-processing. This means that they can use the filters that
have to be built by user in other algorithms. CNNs can be
utilized in various applications from image and video
recognition, image classification, andrecommendersystems
to natural language processing and medical image analysis.
1. Input: This layer have the raw pixel values of image.
2. Convolutional Layer: This layer gets the results of the
neuron layer that is connected to the input regions. We
define the number of filters to be used in this layer. Each
filters that slider over the input data and gets the pixel
element with the utmost intensity as the output.
3. Rectified Linear Unit [ReLU] Layer: This layer applies an
element wise activation function on the image data. We
know that a CNN uses back propagation. Thus in order to
retain the equivalent values of the pixels and not being
modified by the back propagation, we apply the ReLU
function.
4. Pooling Layer: This layer performs a down-sampling
operation along the spatial dimensionsarewidthandheight,
resulting in volume.
5. Fully Connected Layer: This layer is used to compute the
score classes i.e. which class has the maximum score
corresponding to the input digits.

Fig -5: Training and prediction using CNN.
The datasets used are public databases that are utilized for
diagnosis of carcinoma. The most frequently applied data
from these datasets are LIDC-IDRI dataset. Dataset is having
a carcinoma screening and computerizedtomographyscans,
processed with the feature extraction.
Fig -6: Testing phase of CNN
Then the step by step convolutional neural network
operation takes place. Theconfusion matrixispredicted.Fig-
6 shows the testing flow of CNN. The testing phase of CNN
detects whether the cancer nodule is malignant or benign.
4. EXPERIMENTAL RESULTS
The dataset used in this paper is a collection of CT images of
the carcinoma affected persons and also normal persons.
Those images are of DICOM format,everyindividual imageis
having a multiple axial slices of the chest cavity. Those slices
are displayed in the 2d form of slices. All the medical images
are stored in microdicom format. The input image of dicom
format is transformed by converting to .png, bmp and jpg
format. The pydicom package which is available for spyder
environment is used. The python language works good with
all the dicom format images. At valuation several metrics are
utilised. Using confusion matrix, the performance is
calculated. The binary classification technique is also
realized. Confusion matrix is the easily understandable
metrics used to find the model's accuracy. The accuracy of
the system is determined by looking at the TN, TP, FN, and
FP. The Results for the SVM classifiers are shown as various
parameters like confusion matrix, accuracy score, and
reports are extracted. Then followed by receiver operating
characteristic curve is obtained.
Table -1: Confusion matrix of the SVM
4 1
1 4
Accuracy score [0.8]
Table -2: Report for the SVM classifier
classes Precision Recall f1-score support
0.0 0.80 0.80 0.80 5
1.0 0.80 0.80 0.80 5
micro
average
0.80 0.80 0.80 10
macro
average
0.80 0.80 0.80 10
weighted
average
0.80 0.80 0.80 10
Chart -1: ROC curve for SVM
The Results for the CNN classifiers are as shown below
various parameters like confusion matrix, accuracy score,
and reports are extracted and followed by receiver
operating characteristic curve. In the report the precision,
recall,f1-score, support are obtained for both the classes.

The classes are both normal images and tumor images.
Table -3: Confusion matrix of the CNN
6 0
1 3
Accuracy score [0.9]
Table -4: Report for the CNN classifier
classes Precision Recall f1-score support
0.0 0.86 1.00 0.92 6
1.0 1.00 0.75 0.86 4
micro
average
0.90 0.90 0.90 10
macro
average
0.93 0.88 0.89 10
weighted
average
0.91 0.90 0.90 10
Chart -2: ROC curve for CNN
Comparison of SVM and CNN classifier is shown using the
accuracy score and differentparameterslikeprecision,recall,
f1-score,supportparameters.Theaccuracyscorewithrespect
to two different classes and with the two different classifiers
are obtained.
Table -5: Accuracy score
classifier SVM CNN
accuracy 0.8 0.9
Table -5: Comparison of various parameters of SVM and
CNN
REPORT SVM CNN
Classes 0.0 1.0 0.0 1.0
Precision 0.80 0.80 0.86 1.00
Recall 0.80 0.80 1.00 0.75
f1-score 0.80 0.80 0.92 0.86
Support 5 5 6 4
After testing, output is obtained using the SVM classifier and
CNN classifiers to detect whetherthenodulesaremalignancy
or benign. The individual operation for SVM and CNN is
obtained using the CT images as the input.
(a) (c)
(b) (d)
Fig -7: Lung cancer CT scans (a) Input image, (b) Median
filtered image, (c) Nodules representation, (d) Detection of
nodule as normal nodules.
(a) (c)

(b) (d)
Fig -8: Lung cancer CT scans (a) Input image, (b) Median
filtered image, (c) Nodules representation, (d) Detection of
nodule as tumor nodules
5. CONCLUSION
This study draws attention to the diagnosis of lung cancer.
Lung nodule classification is benign and malignant. The
proposed method CNN architecture is specially regarded for
its success in image classification compared to support
vector machine. For biomedical image classification
operation, it alsoobtainssuccessful results.CNN architecture
is used for classification in the study. Experimental results
show that the proposed method is better than the support
vector machine in terms of various parameters. The images
in the data set used are rather small. In the future, the
performance of the system can be improved with a larger
dataset and an improved architecture. The proposedsystem
is able to detect both benign and malignant tumors more
correctly.
ACKNOWLEDGEMENT
We express our thanks to Department of Electronics and
Communication, Bangalore Institute of Technology for
providing their constant support and encouragement.
REFERENCES
[1] Anita Chaudhary, Sonit Sukhraj Singh “Lung cancer
detection on CT images using image processing”, is
computing sciences 2012 international conference,
IEEE, 2012.
[2] G. Guo, S. Z. Li, and K. Chan, “Face recognitionbysupport
vector machines,” In Proceedings of the IEEE
International Conferenceon AutomaticFaceandGesture
Recognition,Grenoble,France,pp.196-201,March2000.
[3] D. Nurtiyasari, and R. Dedi, "The application of Wavelet
Recurrent Neural Network for lung cancer
classification," Science and Technology-Computer
(ICST), 2017 3rd International Conference on. IEEE,
2017.
[4] K. He, X. Zhang, S. Ren, and J. Sun, J. “Deepresidual
learning for image recognition,” In Proceedings of the
IEEE conference on computer vision and pattern
recognition (pp. 770-778). 2016.
[5] E. Dandıl, M. Çakiroğlu, Z. Ekşi, M. Özkan, Ö. K. Kurt, and
A. Canan, “Artificial neural network-based classification
system for lung nodules on computed tomography
scans,” in 6th International Conference of Soft
Computing and Pattern Recognition (soCPar), pp. 382–
386, IEEE, 2014.
[6] J. Cabrera, D. Abigaile and S. Geoffrey, "Lung cancer
classification tool using microarray data and support
vector machines,"Information,Intelligence,Systemsand
Applications (IISA), 2015 6th International Conference
on. IEEE, 2015.
[7] H. M. Orozco and O. O. V. Villegas, “Lung nodule
classification in CT thorax images using support vector
machines,” in 12th Mexican International Conferenceon
Artificial Intelligence, pp. 277–283, IEEE, 2013.
[8] H. Krewer, B. Geiger, L. O. Hall et al., “Effect of texture
features in computer aided diagnosis of pulmonary
nodules in low dose computed tomography,” in
Proceedings of the IEEE InternationalnConference on
Systems, Man, and Cybernetics (SMC),2013, pp. 3887–
3891, IEEE, Manchester, United Kingdom,2013.
[9] M. Abadi, P. Barham, J. Chen, Z. Chen, A.Davis, J. Dean,
M.cDevin, S. Ghemawat, G. Irving, M. Isard, M. Kudlur,
“TensorFlow: A System for Large-Scale Machine
Learning,” InOSDI 2016 Nov 2 (Vol. 16, pp. 265-283).
[10] Arvind Kumar Tiwari, “Prediction Of Lung Cancer Using
ImageProcessingTechniques”,AdvancedComputational
Intelligence: An International Journal (Acii), Vol.3, No.1,
January 2016.
[11] Raquel Kolitski Stasiu, Gerson Linck Bichinho,
“Components Proposal for Medical Images and HIS”,
Proceedings of the Fourteenth IEEE Symposium on
Computer-Based Medical Systems, Page 73, 2001,
ISSN:1063-7125.

IRJET- Lung Cancer Nodules Classification and Detection using SVM and CNN Classifiers

Recommended

More Related Content

What's hot (20)

Similar to IRJET- Lung Cancer Nodules Classification and Detection using SVM and CNN Classifiers (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Lung Cancer Nodules Classification and Detection using SVM and CNN Classifiers