2024 International Conference on Data Science and Network Security (ICDSNS)

Implementation Of Lung Cancer Nodule Detection

Using CT scan
Manoj Kumar D P Deepika G L Sariya Mazhar
Computer Science Department Computer Science Department Computer Science Department
Kalpataru Institute Of Technology Kalpataru Institute Of Technology Kalpataru Institute Of Technology
Tiptur, Karnataka, India – 572201 Tiptur, Karnataka, India – 572201 Tiptur, Karnataka, India – 572201
[email protected] [email protected] [email protected]

Preeti T L Pooja P Namitha C

Computer Science Department Computer Science Department Computer Science Department
Kalpataru Institute Of Technology Kalpataru Institute Of Technology Kalpataru Institute Of Technology
Tiptur, Karnataka, India – 572201 Tiptur, Karnataka, India – 572201 Tiptur, Karnataka, India – 572201
[email protected] [email protected] [email protected]

Abstract— The paper focuses on enhancing the precision in Furthermore, manual analysis of imaging data is time-
detecting and measuring small lung nodules which are crucial consuming and prone to errors, necessitating more reliable
for early diagnosis and treatment of lung cancer. Machine and efficient diagnostic tools. The urgency of early and
learning and image processing techniques are used to overcome accurate lung cancer nodule diagnosis motivates the study and
traditional imaging techniques where the radiologists often hence we are going to overcome all the above challenges.
struggle to distinguish nodules from surrounding tissues due to Initially to extract the Lung cancer nodules, a novel automatic
limited resolution in the images. This study improves small lung methodology has been proposed, which is mainly based on the
nodule detection using advancements in high-resolution Time - black circular neighborhood rule and image processing
0f-Flight, Positron Emission Tomography and resolution
techniques (P.B. Bach et al) [4]. In algorithms, feature
recovery algorithms. Techniques like Otsu thresholding,
Watershed transform, and GLCM were used. Simulations
extraction is implemented. These divide the area, which is
evaluated nodule accuracy for sizes 4-10 mm, contrast levels 2:1
subsequently examined to look for nodules that would indicate
to 8:1, count levels 1%-100%, and respiratory motion the sickness and we have made use of CT imaging and GLCM
amplitudes. Otsu thresholding was used for image options that aid in nodule detection. The application of Otsu's
reconstruction. Watershed transform effectively segmented rule facilitates the determination of the tumor's size and stage
foreground from background. GLCM features from CT scans (W. C. Hanna et al) [5].
enabled accurate lung cancer nodule identification using an
This research paper consists of Introduction, Literature
SVM algorithm. Combining advanced imaging techniques and
Survey, Methodology, Experimental Results and finally the
algorithms significantly improves the accuracy and reliability of
small lung nodule detection, aiding early lung cancer
Conclusion
diagnosis and treatment.

Keywords— Otsu thresholding technique, GLCM feature,

Watershed, SVM Machine Learning Algorithm.

I. INTRODUCTION
The lungs are a pair of cone shaped, sponge-like organs.[1]
The right lung has three lobes, and it is larger than the left
lung, which has two lobes. Anatomy of lung is shown in Fig.1.
Lung cancer is a disease where cells multiply abnormally and Figure1: Anatomy of Lung
grow into a nodule. Fig.2 describes the beginning of the
cancer. There are four stages of lung cancer. In stage I, the
cancer is limited to the lung. In stages II and III, the cancer
infects the chest (with larger and more invasive tumor
classified as stage III). In the stage IV, the cancer is in both
lungs or has spread to fluid around the lungs or other parts of
the body.[1].
Despite advancements in medical imaging and
diagnostics, early detection of lung cancer remains
challenging. Existing studies highlight difficulties in
identifying small nodules, distinguishing between benign and Figure 2:Beginning of Cancer
malignant growths, and accurately staging cancer using
conventional methods (Wang-Jia Li et al, 2022) [3].

979-8-3503-7311-0/24/$31.00 ©2024 IEEE

 II. LITERATURE SURVEY Year:2017
Paper 1: “A morphological operation-based approach for Explanation: This work suggests the use of 3D
Subpleural lung nodule detection from CT image” convolutional neural networks for the detection of lung
Authors: Rekka Mastouri, Henda Neji, Saoussen nodules in lung CT scans. Lung CT images are subjected
Hantous- Zannad, Nawres Khlifa to a combination of conventional morphological pre-
processing techniques and 3D convolutional neural
Year: 2018 networks.
Explanation: This study focuses on a method for
automatically segmenting sub-pleural lung nodules from CT Paper 5:” Segmentation and Prediction from CT Images
scans that are dependent on morphological processes. Due to for Detecting Lung Cancer”
the difficulty in extracting sub-pleural nodules, a computer-
aided diagnosis system is therefore essential. Three processes Authors: K.S Chethan, S. Vishwanath, rakshith V.Patil,
make up the suggested system: pre- processing, initial sub- K.A Vijetha
pleural lung nodule detection, and post-processing
Year: 2020
Paper 2: “A survey on detection of lung cancer using
different image processing techniques” Explanation: Throughout the world, cancer is among the most
common medical conditions. Lung cancer is the most
Authors: Sanjana Narvekar, Mayur Shirodhar, Tanvi Raut, common type of cancer among the general population, though
Purva Vaingankar, k. M.chaman Kumar, Shailendra Aswale there are other varieties as well. One of the most deadly
medical conditions worldwide, lung cancer affects people of
Year: 2020 all genders. One crucial measure that can reduce the chance of
Explanation: The world is changing at an incredible rate, and death from cancer is early detection. Our work performs lung
with it, the number of cases of malignant lung cancer nodule categorization by utilizing the CT scan data set of the
development. One such instance is cellular breakdown in the lungs that was gathered from the websites of Kaggle in early
lungs, which has claimed the lives of many people. Positron stage. Lung nodule categorization has become more and more
emission tomography (PET), magnetic resonance imaging dependent machine learning algorithms in recent years.
(MRI), X-ray imaging, magnetic resonance imaging (MRI),
and computed tomography (CT) scans are used to detect III. . METHODOLOGY
cellular breakdown in the lungs. Comparing X-ray and PET 1. Architecture Diagram
sweeps to CT and X-ray, the former are more expensive.
Compared to other imaging procedures, CT photos are
preferred. The great majority of specialists choose X-rays or
CT sweeps. By applying image management techniques to the
image, we may decompose CT sweeps and X-ray images with
greater depths to prevent such irregularities. CT.
Paper 3: “Lung Cancer Screening Using Computer-Aided
Detection and Image Processing Techniques”
Authors:Anil Boddupalli; Laxman Rayala; Sai Sandeep
Lingareddy; Yetra Mohan Sai Saran Reddy; Yamuna Devi M
M, A. Veeraswamy
Figure3: System Architecture
Year: 2023
Explanation: Lung cancer accounts for 11% of all deaths in The lung picture is used as an input in this system, and
India, making it one of the top causes of death. Finding the several approaches are applied to identify the lung nodule.
area affected by lung cancer has been easier recently because Using Otsu's thresholding method, we first calculate a
to advances in image processing techniques. Better treatment measure of spread for the pixel levels on each side of the
outcomes for lung cancer may arise from early detection. It is threshold—that is, the pixels that fall into the foreground or
much preferred to use applied image processing techniques to background—by iterating over all conceivable threshold
evaluate CT scan data and identify lung illnesses in humans. values. The watershed algorithm is used to filter images in
This can also help in identifying the problem regions in a preparation for image segmentation. The watershed transform
number of ways, making it possible for people to receive the is a better segmentation technique for identifying foreground
right kind of healthcare support. Here, the infected lung objects and background locations. by using the GLCM
region has been clearly identified with the application of feature, which is computed from the CT image's discovered
classification algorithms such as Support Vector Machines lung nodule. Lastly, we use the SVM machine learning
(SVM). Further research findings on pre-processing, technique to identify lung nodules.
segmentation, and feature extraction are presented in this 2. Dataset
work
Datasets used in lung cancer nodule detection using CT
Paper 4: “Lung nodule detection based on 3D convolution
scans typically contain annotated medical images that are
neural networks” crucial for developing and validating algorithms for
Authors: Lei Fan, Zhaoqiang Xia, Xiaobiao Zhang, automated detection. CT Scan Images: The dataset consists of
Xiaoy Feng CT (Computed Tomography) scan images of the chest area of
patients. These images are typically in DICOM (Digital
Imaging and Communications in Medicine) format, which is
standard in medical imaging. 2)Watershed Algorithm:

Annotations: Each CT scan is annotated to mark the

presence and location of lung nodules. Annotations are
usually provided by radiologists or medical experts who
have identified and categorized the nodules based on their
characteristics such as size, shape, and density.
Variety of Nodules: The dataset may include various types
of nodules (e.g., solid, ground-glass opacity, part-solid)
and nodules of different sizes, as these characteristics
influence detection algorithms.
Size and Scope: The size of the dataset can vary widely
depending on its purpose. Some datasets may include
hundreds to thousands of CT scans, while others may be
smaller and more focused.
3)SVM Algorithm:
DatasetLink:https://www.kaggle.com/datasets/sariyamaz
har/lung-cancer-nodule-detection-dataset[12]
C.Libraries

i) OpenCV

OpenCV (Open-Source Computer Vision Library) is an

open- source computer vision and machine learning software
library
i) Flask

A compact and light-weight Python web framework, Flask

offers practical tools and capabilities that facilitate the 4)GLCM Algorithm:
development of Python web applications
i) Keras
An open-source package called Keras offers an artificial
neural network Python interface
iv)Matplotlib
For the Python programming language and its NumPy
numerical mathematics extension, Matplotlib is a graphing
library
v)NumPy
A Python package called NumPy is used to work with arrays.
vi)TensorFlow
TensorFlow is an open-source machine learning platform and
framework.
D.Algorithms:
1)Otsu Threshold Algorithm:
E. Implementation:
1. Data Flow Diagrams

Figure:4: Data Flow Diagrams 0

 The project parameters can be roughly estimated using the
basic COCOMO model. The following expressions yield the
basic COCOMO estimate model:
Effort = a1*(KLOC)a2PM
Tdev = b1 x (Effort)b2 Months
Figure 5 : Data Flow Diagrams 1
Where,
1) KLOC is the estimated size of the software product
expressed in Kilo Lines of Code,
2) a1, a2, b1, b2 are constants for each category of
software products,
3) Tdev is the estimated time to develop the software,
expressed in months,
4) Effort is the total effort required to develop the
software product, expressed in person months (PMs).
The effort estimation is expressed in units of person-
months (PM). It is the area under the person-month plot. It
should be carefully noted that an effort of 100 PM does not
imply that 100 persons should work for 1 month nor does
Figure 6: Data Flow Diagrams 2 it imply that 1 person should be employed for 100 months,
but it denotes the area under the person-month curve.
2. Software’s Used ii. Waterfall Model:
i. Python compiler [3.6.11]
An application called a Python compiler transforms your
human-readable Python code into a lower-level language that
the computer's hardware can execute directly, usually
bytecode or machine code
ii. Visual Studio Code:
Microsoft created Visual Studio Code, popularly known
as VS Code, which is a source-code editor compatible with
Windows, Linux, macOS, and web browsers. Debugging,
syntax highlighting, intelligent code completion, snippets,
code refactoring, and integrated Git version control are
among the features.
iii. Anaconda environment:
A directory containing a particular set of installed Conda
packages is called a Conda environment.
iv. Installing required packages: 4. Proposed System:
Here, we're using OpenCV 4.10.0 + and the Python In this system, we have a tendency to use respiratory organ
compiler version 3.6.11+. Moreover, SciPy and NumPy are image as an input and apply some techniques to spot the
utilized for certain computations needed for this experiment. nodule of the respiratory organ. Here 1st we have a tendency
➢ The core Python library for large-scale to use Otsu's thresholding methodology involves iterating
through all the potential threshold values and shrewd a life of
mathematical computation is called NumPy.[13]
unfolding for the pel levels either side of the edge, i.e. the
➢ SciPy is an open-source, BSD-licensed scientific pixels that either fall in foreground or background. then that
library for Python that is used in science, engineering, and pictures are filtered for image segmentation by victimization
mathematics computation applications. [13] watershed formula.
3. Models used: Segmentation victimization the watershed rework works
higher to establish, foreground objects and background
locations. By applying the GLCM feature that reason from the
i. COCOMO Model: detected respiratory organ nodule in the CT image. and
Boehm proposed the COCOMO (Constructive Cost eventually, we have a tendency to apply the SVM machine
Estimation Model) [1981]. Boehm suggests three steps for learning formula for detective work nodule of
software cost estimation: Basic COCOMO, Intermediate respiratory organ.
COCOMO and Complete COCOMO.
 IV. EXPERIMENTAL RESULTS The procedures and methods shown in the snapshot are
thoroughly summarized in this description.(i) The user first
uploads the original grayscale lung CT scan image, which
displays the lungs' cross-sectional view.(ii) Using Otsu's
thresholding technique, our system computes Otsu's Binary
Threshold, a binarized version of the input image that is used
to differentiate between various regions by turning the image
into black and white.(iii) A picture that has undergone
morphological closing operation (with a 2x2 kernel), which
aids in eradicating tiny black holes from the white areas.(iv)
The image has now undergone dilatation, which enlarges the
limits of the white zone and highlights its features.(v)Step for
the computation Distance Conversion An altered picture that
Figure 7: Home page of Application displays each pixel's separation from the closest zero pixel,
usually utilized.

Figure 8: Receiving inputs from user

The figure 7 and figure 8 depicts a graphical user interface

(GUI) of a software application titled "Lung Cancer Nodule
Feature Extraction Using Digital Image Processing System." Figure 9: The feature extraction of the lung by GLCM
The left side of the screen has a number of buttons in a
vertical alignment inside the interface. The buttons are
labelled as follows, top to bottom:
*Uploading a Lung CT image
*Image Processing and Segmentation
*Feature Extraction
*Application for Histogram Classification
*Exit
The UI has a subdued blue background. The GUI
elements and their arrangement in the screenshot are
succinctly and clearly summarized in this explanation.

Figure 10: Histogram

This plot is a histogram of pixel intensities for a given image.

The high peak at the low intensity values suggests that the
image contains a large number of dark pixels. The subsequent
low and relatively flat line indicates fewer occurrences of
mid-range and high-intensity pixels.
An explanation of the storyline:
Y-Axis: 0 to 70,000 is the vertical axis. The frequency or
count of pixel values in the image is most frequently
represented by this axis.
Figure 9: Image Processing X-Axis: 0 to 255 is the range of the horizontal axis. The
grayscale values, or pixel intensity levels, of an image are
generally represented by this axis.
Features of the Storyline:
A high frequency of pixels with low intensity values is shown B. HARDWARE REQUIREMENTS
by the plot's abrupt peak at the 0-intensity value at the  Hardware : Pentium Dual Core
beginning. As the intensity values rise, the frequency sharply  Speed : 2.80 GHz
decreases after the initial peak and stays quite low with just
slight variations. Intensity levels range from mid to high, with  RAM : 1GB
a few minor peaks and troughs; however, these are much less  Hard Disk: 20 GB
than the first peak.
VI. CONCLUSION

Our work highlights the viability of using low-dose PET

imaging to quantify sub-centimeter nodules, with the potential
to achieve measurement errors of no more than 20% based on
simulation results. However, comprehensive validation
utilizing clinically relevant phantoms is necessary to achieve
substantial clinical application. To improve accuracy, it is
crucial to follow recommendations such using respiratory
motion correction techniques and using a reconstruction voxel
size of 1 mm for tiny nodules. Further emphasizing the
significance of thorough validation in honing these approaches
for clinical use, the integration of image analysis techniques
like Otsu's thresholding, Watershed segmentation, GLCM
texture analysis, and SVM classification holds promise for
attaining accurate nodule size measurements. It will take
ongoing investigation and validation to improve these methods
Figure 11: Malignant Cancer and increase their applicability in clinical settings.
VII. DISCUSSION
First we upload the lung image to the system typically from
medical imaging techniques such as X-rays, CT scan. Once
the image is uploaded successfully pre-processing is
performed using Otsu. Then we perform segmentation using
Watershed algorithm followed by feature extraction which is
done using GLCM. To obtain the classification of the image
where the output result is either benign or malignant tumor
we used SVM algorithm. This structured approach not only
enhances accuracy but also ensures that each image is
thoroughly examined, making the diagnostic process more
reliable. By leveraging these advanced techniques, healthcare
professionals can make better-informed decisions and to
automate the analysis of lung images to aid in the diagnosis
Figure 15: Benign cancer of lung tumors, ultimately improving patient outcomes.
Then the application will classify whether the cancer is
malignant or benign using SVM. REFERENCES
Hence, we use machine learning approaches to detect [1] World Health Organization (2023) Lung Cancer.
lung nodules. Otsu thresholding was utilized for https://www.who.int/news-room/fact-sheets/detail/lung-cancer
preprocessing, Watershed was used for segmentation, [2] SanjanaNarvekar, MayurShirodhar, TanviRaut, PurvaVaingankar, k.
M.chaman Kumar, ShailendraAswale: “A survey on dection of lung cancer
GLCM was employed for feature extraction, and SVM using different image procseesingtechniques”in year 2020
classification was utilized to obtain the final output.
[3] Wang-Jia Li, Fa-Jin Lv, Yi-Wen Tan, 2 Bin-Jie Fu, and Zhi-Gang Chu-
The study's findings demonstrate that combining these “Benign and malignant pulmonary part-solid nodules differentiation via
advanced imaging techniques and algorithms significantly thin-section computed tomography” Quant Imaging Med Surg. 2022 Jan;
improves the accuracy and reliability of small lung nodule 12(1): 699–710. doi: 10.21037/qims-21-145
detection, aiding early diagnosis and treatment of lung [4] P. B. Bach et al., “Benefits and harms of CT screening for lung cancer: A
cancer. systematic review”, J. Amer.Med.Assoc.,vol.307,no.22,pp.2418–
2429,Jun.2012.
V. SYSTEM REQUIREMENTS [5] W. C. Hanna et al., “Minimal-dose computed tomography is superior to
chest X-ray for the follow-up and treatment of patients with resected lung
A. SOFTWARE REQUIREMENTS cancer,” J. ThoracicCardiovasc.Surgery,vol.147,no.1,pp.30–35, Jan.2014.
 Operating System : Windows 7 and above [6] Anil Boddupalli; LaxmanRayala; SaiSandeepLingareddy; Yetra Mohan
Sai Saran Reddy; Yamuna Devi M M, A. Veeraswamy “Lung Cancer
 Technology : Python
Screening Using Computer-Aided Detection and Image Processing
 IDE : Anaconda or Spider Techniques” in year 2023
[7] Rekka Mastouri, Henda Neji, Saoussen Hantous- Zannad,Nawres Khlifa, ”
A morpho logical operation-based approach for Subpleural lung nodule
detection from CT images ”, 4th Middle East Conference on Biomedical
Engineering (MECBME), IEEE, 2018.
[8] D.R. Aberle et al., “Reduced lung-cancer mortality withlow-dose
computed tomographic screening,”New England J. Med., vol. 365, no. 5,
pp. 395–409, Aug. 2011.
[9] D.Shlomi,R.Ben Avi,G.R.Balmor,A.Onn,and N.Peled,“Screeningforlungc
ancer:Time for large-scale screening by chest computed tomography,” Eur.
Respiratory J., vol. 44, no. 1, pp. 217–238, Jul. 2014.
[10] M. Infante et al., “Lung cancer screening withspiral CT: Baseline results
of the randomized DANTE trial,” Lung Cancer, vol. 59, no. 3, pp. 355–
363, Mar. 2008.
[11] Wikipedia-Informational Contents.
[12] Kaggle Dataset
[13] ]Python 3.11.3 documentation– Virtual environment [Venv] - Creation of
virtual environments
[14] Google-Images
[15] D. S. Gierada et al., “Projected outcomes usingdifferent nodule sizes to
define a positive CT lung cancer screening examination,” J. Nat. Cancer
Inst., vol. 106, no. 11, 2014, Art.no.dju284,doi:10.1093/jnci/dju28