A Comparative study of Lung Cancer Detection and
Classification approaches in CT images
Preeti Katiyar
Department of Electronics and Communication Engineering
Guru Gobind Singh Indraprastha University
New Delhi, India
preetikatiyar27@gmail.com
Krishna Singh
Department of Electronics and Communication Engineering
G.B Pant Govt. Engineering College
New Delhi, India
singhkrishna5@gmail.com
Abstract – Lung disease is a genuine medical problem. In
India there are roughly 70,275 individual cases in every year are
determined to have lung malignancy. However, early
identification and treatment can increase the survival rate.
Usually Computed Tomography scan imaging is used in the
medical field because of its high clarity and low noise. Only CT
scans cannot give proper interpretation to radiologist and the
medical practitioner, therefore the Computer Aided Diagnosis
system will be extremely useful for radiologists to detect the
cancer precisely. Many Computer aided system using image
processing and Machine learning has been designed. In this
survey various segmentation, feature extraction and classification
techniques are considered such as Artificial Neural Network,
Convolutional Neural Network, SVM, Gray level co-occurrence
matrix, Discrete wavelet transform and many more. We observed
that the SVM classifier achieved 96% accuracy, ANN achieved
99% accuracy, CNN achieved 94% accuracy and DNN achieved
97% accuracy.
Keywords– SVM, CNN, GLCM, CT Scan
I. INTRODUCTION
Lung cancer is the second most diagnosed disease among a
various kind of cancers which is available in the world. In
ladies, bosom malignancy is normal and in men, prostate
malignancy is normal. Lung malignancy is one of the most
severe cancer, with the little survival rate after the finding. It is
the most important reason for malignancy demise in both
women and men these days. Generally rate of smoking results
in lung cancer which include 75% females and 84% males.
Around 10-15% instances happen in individuals who never
smoked. Such instances are results of blend of hereditary
variables and presentation of gas like radon, different types of
contaminated air, active and passive smoking and minerals
like asbestos. Early detection of the cancer results in the
survival of the patient. So it is necessary to find out the lung
cancer in the starting phase.
To find the irregularity in an image time factor is very crucial,
particularly in different kind of cancers. For example, lung
malignancy, bosom growth etc. Subsequently, different image
Processing techniques are utilized in different stage of
treatment for advancement in image for early identification.
There are various approaches which are utilized to analyse the
lung nodule such as Computed Tomography, Positron
Emission Tomography and X-ray. Most of these approaches
are costly and needs lots of processing time. So we require a
new computer aided design system to identify the lung cancer
in its initial stage.
In medical radiology, Computer Aided Design System plays
an important role. But most of the system does not fulfil the
requirement of radiologist. It was described that by using
Computed Tomography scans in early stages for the
identification of lung tumor, the survival rate of the patient can
reach upto 90%. So we usually use CT Scans for CAD System
[1]. This CT image is collected from Cancer Image Archive. It
works on cancer analysis and has a large database of medical
imaging [2]. Fig 1 shows CT Scan of Lung Cancer.
Fig.1. CT Scan image of Lung Cancer
In general study the lung cancer diagnosis involves many
different technologies. The CAD systems are developed for
identification of lung cancer in initial stage. A general
Computer Aided Diagnosis set-up consists of several steps in
identification of the lung cancer. The following techniques are
1) Pre-processing and segmentation 2) Nodule Detection 3)
2020 7th International Conference on Signal Processing and Integrated Networks (SPIN)
978-1-7281-5475-6/20/$31.00 ©2020 IEEE 135
Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 30,2020 at 15:50:48 UTC from IEEE Xplore. Restrictions apply.