Special Conference Issue: National Conference on

Cloud Computing & Big Data 104

Segmentation of Lung Nodule Using Image

Processing Techniques from CT Images
Santosh Singh
Research Scholar, Banasthali University, Tonk, Rajasthan
[email protected]
Ritu Vijay
Reader, Department of Electronics, AIM & ACT, Banasthali University, Rajasthan (INDIA)
[email protected]
Yogesh Singh
Department of Physiology, SMS Medical College, Jaipur, Rajasthan

-------------------------------------------------------------------ABSTRACT-------------------------------------------------------------
Lung cancer is the most prevalent and hazardous disease in the world having highest mortality rate among all other
types of cancers like breast, liver, brain, skin etc.. There is a great need to detect it in its early stage to provide a
great chance of survival to the patient. It is possible only when lung nodules will be automatically segmented
through image modalities. The main aim of this study is to overcome the above problem from CT images.

The medical image segmentation with accuracy is a prognostic factor in the diagnosis of lung cancer. Manual
segmentation of lung nodules form CT image is a challenging & crucial task. A semi-automated segmentation
method is proposed to segment lung nodules from CT images which can successfully segment a nodule provided
that the parameters are set properly.

Keywords – Cancer, CT Images, Image modalities, Lung Nodule, Medical Image Segmentation
-----------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Introduction In this study we proposed that preprocessing and
segmentation of image is a more effective technique to
L ung cancer is the most prevalent and hazardous disease diagnose the disease in its early stage that would enhance
in the world having highest mortality rate among all other the quality of the diagnosis speedy.
types of cancers like breast, liver, brain, skin etc. Today, a
number of techniques are available to diagnose the lung By keeping the goals of CAD [7] in mind when achieving
disease, such as X- ray, CT, MRI Scan, PET-CT etc. but the results, segmentation of the image seems to be
majority of death occurs only due to diagnosis in later necessary. Segmentation is described as a method of
stages. dispensing image pixels into meaningful, usually
connected, regions which are homogenous with respect to
Nowadays for lung cancers, computed tomography is more some criterion. In various image processing solutions, it is
reliable and the most effective image modality among all a vital staircase. Segmentation is of chief importance in
modalities due to high accuracy and sensitivity in ruling out medical imaging for image measurements, feature
size of tumor and lymph node metastases [6]. X-rays extraction, and display of image. Various applications it
highlights vital parts of the body, whereas CT serve may be useful to classify image pixels into anatomical
thorough outlook of the soft tissues, including lungs, regions of body. Although, algorithms and segmentation
muscle tissue and blood vessels. While Roentgenogram techniques have been proposed and implemented, the
provides a 2-D image of body, a sliced view of the body is problem remains unsolved. The higher efficient lung
depicted by CT images. segmentation helps in analysis and diagnosis for lung
diseases [3].
Advantages and disadvantages of CT images are
acknowledged by Sharma N. et al in his study [7]. Despite Techniques of image segmentation can be grouped into
the disadvantages, CT modality plays an important part in three based on the following parameters
the evaluation of lung nodule, tuberculosis, pneumonia, Intensity (example thresholding)
emphysema, inflammation etc. Region (example split and merge and region growing)
Other methods (example edge, motion and texture
As detection of the lung nodules from CT images is more based segmentation)
challenging, there would be a need to develop a more
reliable and an effective CAD system to detect the disease 2. Methodology
in its early stage, thus increasing the possibility of survival The preprocessing steps of the proposed system can
to the patients. [2] identify the appropriate region affected by cancer. The aim
of preprocessing is to get better quality of the acquired
Special Conference Issue: National Conference on
Cloud Computing & Big Data 105

image. Preprocessing involves image enhancement, noise achieved. The goal of histogram equalization is to provide
removal, thresholding, segmentation. Every progression is uniformity which facilitates further visually satisfying
successfully completed by using MATLAB software which results. Mathematically, each individual histogram can be
provides a multi-platform environment for the user. This expressed as
tool is used for prototyping, analysing the data, and h(k) = nk = card{(x , y) | f(x , y) =k } (1)
provides a built-in support for various operations. Also, it
has an enriched capability of graphics with a user friendly Here, k = 0,1,…, L-1, where L is the number of gray levels
programming language environment. In the software of the digitized image, and card(…) denotes the cardinality
development cycle, it leads to precious time savings. of a set, that is, the number of elements in that set(n k).

2.1 Preprocessing Mathematically, a normalized histogram can be defined as

At first we convert the image to gray scale [3] image that p(rk) = nk
contain only brightness information. These images are n
determined as a two dimensional array of pixels which is
using 8 bits/pixels. Here, zero pixel value correspond to Where n is the total number of pixels in the image and p(r k)
black and white is for 255, with transitional values is the probability (percentage) of the kth gray level (rk).
corresponding to varying shades of gray. Conversion of
image into gray scale provides an advantage of reducing Figure (c) represents the histogram of the original image
the processing time facilitating rapid algorithm with one bar per gray level in which the height of the bar is
production[6]. proportional to the number (or percentage) of pixels that
correspond to that particular gray level. Figure (d) signify
the uniform histogram after equalization using imhist
function.

(a) Original image (b) Grayscale image

Often the quality of the acquired image is not satisfactory

due to factors such as blur, unnatural colors, noise, and
artifacts. It is necessary to pre-process the image so that the
irrelevant information or noise is removed. This process is
called Image Enhancement i.e. it makes the image more
useful for further processing. The reasons for doing this
consist of highlighting interesting details, removing noise
and making images more visually appearing.
Figure (c) Histogram of original image
To enhance the quality of an image, it is necessary to assess
its features. This requires some quality quantification &
assessment tools. Enhancement techniques can be grouped
into two broad classes as frequency domain and spatial
domain. Spatial domain includes direct manipulation of
image pixels, while manipulation of fourier transform or
wavelet transform of an image comes under frequency
domain.

For the moment, this study concentrated on techniques that

operate in the spatial domain i.e. Histogram Equalization.

2.2 Histogram Equalization

The frequency of occurrence of each gray level in the Figure (d) Histogram Equalization
image is represented by a graphical representation known
as histogram of a monochrome image. It reassigns the 2.3 Noise Removal
brightness values of pixels based on the image histogram. Apart from contrast and intensity manipulation image
Histogram- modeling techniques are used to modify the enhancement includes noise removal, filtering etc which
image so that its histogram has a desired shape. This is make the image suitable for further processing [2]. Image
useful in stretching the low-contrast levels of images with smoothing is useful for giving a softer visual effect and for
narrow histograms. Histogram modeling has been found to
removing noise. Noise can be defined as any unwanted
be a powerful technique because they are simple, fast and
with them acceptable results for some applications can be information that contaminates an image. The presence of
Special Conference Issue: National Conference on
Cloud Computing & Big Data 106

noise in an image can be due to several sources like analog- The best move towards optimal thresholding was projected
to-digital conversion, camera sensors, communication by Otsu and implemented by graythresh which compute
channels etc. resulting in different types of noises like global threshold. This method is based on threshold
Gaussian selection criteria. Based on this method, threshold value
Impulse (Salt and pepper) will be between 0 and 1. After achieving this value we can
Rayleigh segment an image based on it [1].
Speckle
Uniform
Exponential
Gamma(Erlang)

Enhancement of the acquired image focus on removing

noise that is the product of low-level data errors. Therefore,
it is vital to remove noise to improve the quality of the
acquired image and to facilitate the further processing such Figure (f) Threshold Image
as edge detection, segmentation and analysis etc. Filtering
an image to smoothen noise while maintaining the details 2.4 Morphological operations
The main morphological operations are dilation, erosion,
of the image preserved is one of the most important issues.
closing, opening and the hit-or-miss transform. Among
Median Filter is the most admired filter among all others these apply morphology closing (dilation and erosion) on
types. It is a nonlinear common enhancement digital filter the image. It fills the indentation caused by the pulmonary
[4]. It selects the middle pixel value form the ordered set of vessels [5]. The effect of dilation is to “grow” or “thicken”
values within the m x n neighbourhood (W) in the region of objects in a binary image whereas, in erosion, outcome is to
the reference pixel. Median filter is advantageous from “shrink” or “thin” objects. The extent and direction of the
smoothing filters as it provides: thickening and thinning are controlled by the shape and
No reduction in contrast across steps size of the structuring element. In the present study
No shifting of boundaries (no disturbance in edges) as structuring element is disk of radius is one. Dilation is
well as does not degrade edges, implemented by using function imdilate() and erosion by
An effective way for smoothing additive white noise in imerode().
removing impulses.

Figure (g) Eroded Image

Figure (e) Filtered image

Thresholding
The purpose of the lung segmentation required for the CAD
from CT scan images is to essentially separate the voxels
corresponding to the lung cavity in the axial CT scan slices
from the surrounding lung anatomy. An optimal
thresholding which selects the threshold based on the
object and background pixel means. Figure (h) Dilated Image
The simplest method within the intensity based methods is 3. Conclusion
image thresholding. Because of its simplicity, intuitive Lung cancer is the widespread disease in the world which
properties and ease of implementation it has become a well is most dangerous and have highest mortality rate among
accepted method. In machine visual systems, thresholding all other types of cancers. The aim of this study is to
is a common preprocessing step. The fundamental setback provide systematic and quantitative measurement on CT
of thresholding is the alteration of the image with several images so that physicians can diagnose the nodule more
gray levels into another image with smaller quantities of effectively and more accurately. By this work, the acquired
results are satisfactory to detect the nodule as mentioned
gray levels, usually two only.
above. This method has improved the efficiency of the
CAD system after applying the preprocessing on the image.
Special Conference Issue: National Conference on
Cloud Computing & Big Data 107

References
[1] Hashemi A, Pilevar A H, Rafeh R, “Mass Detection in
Lung CT Images Using Region Growing Segmentation
and Decision Making Based on Fuzzy Inference
System and Artificial Neural Network”, International
Journal of Image, Graphics and Signal Processing
(IJIGSP), 2013, 6, pp 16-24.
[2] Mesanovic N, Grgic M, Huseinagic H, Males M,
Skejic E, Smajlovic M, “Automatic CT Image
Segmentation of the Lungs with Region Growing
Algorithm”, Proc. 18th International Conference on
Systems, Signals and Image Processing, IWSSIP 2011,
pp 395-400.
[3] Sharma N, Aggrawal L M, “Automated medical image
segmentation techniques”, Journal of Medical Physics,
Jan-Mar 2010, vol 35(1), pp 3-14.
[4] Kumar S, Kumar A, “Lung Segmentation Using
Region Growing Algorithm ”, International Journal of
Advanced Research in Computer Science and Software
Engineering, vol. 4, Issue 3, March 2014, pp 184-187.
[5] Kumar V, Saini A, “Detection system for lung cancer
based on neural network: X-Ray validation
performance ”, International Journal of Enhanced
Research in Management & Computer Applications,
vol. 2, Issue 9, Nov - Dec 2013, pp 40-47.
[6] Chaudhary A, Singh S S, “Lung Cancer Detection
Using Digital Image Processing”, International
Journal of Research in Engineering & Applied
Sciences, vol 2, Issue 2, February 2012, pp 1351-1359.
[7] Lin D, Yan C, “Lung nodules identification rules
extraction with neural fuzzy network”, Proc. 9th
International Conference on Neural Information
Processing (ICONIP 02), vol 4.