FACULTY OF SCIENCE AND TECHNOLOGY

DEPARTMENT OF COMPUTER SCIENCE

NAME: MUPIKATA PRAYER MUNASHE

REG NUMBER: R182254V

LEVEL: 4.1

MODE OF ENTRY: Conventional

YEAR: 2021

PROGRAM: BSc COMPUTER SYSTEMS ENGINEERING

TOPIC: Lung Cancer Detection and Classification Using

Digital Image Processing

1.1 INTRODUCTION

The purpose of this research is to help doctors detect lung cancer using smartphone camera.
Using an image classifier model, we plan on helping doctors in all the detect diseases on their
lungs and get the right medications to treat the patient.

The artificial intelligence model runs on the device CPU which means there will be no need for
broadband data for doctors to use the application, which will help them in cutting unnecessary
costs.

1.2 BACKGROUND OF THE STUDY

Cancer remain a major threat to the humans if not noticed at its early stages across the world, but
their identification is always difficult due to lack of necessary technologies to identify them.  The
combination of increasing global smartphone penetration and recent advances in computer vision
made possible by deep learning has paved the way for smartphone-assisted disease diagnosis.

Lung cancer represents a spectrum of biologically different tumours . These include very
aggressive entities with rapid growth (tumour volume doubling within weeks) and spread of
metastases within weeks or months such as small cell lung cancer (SCLC) and some subgroups
of non-small cell lung cancer (NSCLC) such as poorly differentiated large cell lung cancer. On
the other hand, other tumours take years to double their volume and develop lymphatic or
haematogenous metastases, e.g. bronchioloalveolar carcinoma and other types of well-
differentiated adenocarcinoma. Nevertheless, lung cancer is the leading cause of death from
malignancy worldwide with an estimated 1.3 million deaths per year. Despite extensive research
and improvement in surgical, oncologic and radiation therapy during the last decades, its
prognosis remains dismal with an overall 5-year survival of <15%. This is predominantly due to
the fact that, in most patients, the diagnosis is made at advanced stages either with infiltration of
adjacent structures or with lymphatic or distant metastases. However, if the diagnosis is made at
an early stage with no metastases, 5-year survival in NSCLC may be >65% and with very small
lesions as high as >80%. For SCLC, the difference in cure rates at early versus advanced tumour
stages, unfortunately, is much less pronounced. However, SCLC currently accounts for only
approximately 20% of lung cancer cases. Therefore, hope for improved cure of lung cancer
patients was based on approaches to detect small and usually asymptomatic early stages non-
small cell lung cancer using diagnostic tests.

In more recent times, such efforts have additionally been supported by providing information for
disease diagnosis online, leveraging the increasing Internet penetration worldwide. Even more
recently, tools based on mobile phones have proliferated, taking advantage of the historically
unparalleled rapid uptake of mobile phone technology in all parts of the world (ITU, 2015).
Smartphones in particular offer very novel approaches to help identify diseases because of their
computing power, high-resolution displays, and extensive built-in sets of accessories, such as
advanced HD cameras. The combined factors of widespread smartphone penetration, HD
cameras, and high-performance processors in mobile devices lead to a situation where disease
diagnosis based on automated image recognition, if technically feasible, can be made available at
an unprecedented scale.

Computer vision, and object recognition in particular, has made tremendous advances in the past
few years.  While training large neural networks can be very time-consuming, the trained models
can classify images very quickly, which makes them also suitable for consumer applications on
smartphones. In order to develop accurate image classifiers for the purposes of lung cancer
diagnosis, we needed a large, verified dataset of images of diseased and healthy lungs.

1.3 LITERATURE REVIEWS

Chest radiography and sputum cytology were extensively analyzed for their potential as
screening instruments in smokers in the 1970s. However, results were disappointing in that more
tumours were diagnosed and resected, but mortality from lung cancer did not decrease in the
screening groups compared with groups with no screening. The explanation is probably that both
tests are not sensitive enough to detect early cancers.

In the 1990s, CT was shown to be much more sensitive for small pulmonary nodules, which is
the most common presentation of NSCLC. Unfortunately, standard chest CT is associated with a
relatively high level of radiation exposure, which in a screening setting is particularly
problematic and important as it affects a large proportion of individuals without lung cancer. It
has, however, been shown that the sensitivity and specificity of CT for small lung cancers did not
decrease significantly with major reduction of radiation exposure.
Recently non-radiological tests have been proposed to screen for lung cancer. Fibreoptic
bronchoscopy, particularly when using fluorescence or autofluorescence, is able to identify early
cancer or even precancerous lesions, however, there are several problems which as yet prevent
its widespread use in a population at risk. It is semi-invasive, is not generally accepted by
asymptomatic individuals, is not widely available, and is too expensive and time consuming. It
may, however, be useful to identify and even treat early lesions if the affected individuals are
diagnosed with another test. Laboratory tests using modern techniques may allow the diagnosis
of lung cancer to be made non-invasively in the future. Different approaches for testing for
specific molecular markers in sputum, serum or exhaled air, assessing genetic anomalies in
sputum or blood cells and other tests are under investigation.

1.4 PROBLEM DEFINATION

It is absolutely clear, that the best way to prevent death from lung cancer is for a person to never
smoke and if started, to give up smoking immediately. However, unfortunately, the success rates
of all smoking prevention and cessation programs are poor. If a screening program is to be
instituted, effects on smoking habits need to be assessed. Both positive and negative effects may
be possible. If the invitation to a screening test could increase awareness of the serious health
effects of smoking, potentially supported by demonstrating the extent of smoking-related non-
tumorous changes (e.g. emphysema, atherosclerotic disease), the test in combination with a
smoking cessation program could prove doubly beneficial. If, however, the screened person
(erroneously) believes, that there is no need to stop smoking as the test “will identify any cancer
in time to cure it and there is, therefore, no need to stop smoking” the test could increase the
likelihood of dying from lung cancer instead of decreasing it. This process is inconclusive and
has many barriers such as expensive broadband data and too many search results that the patients
won’t understand.

As a result, I am proposing a lung cancer detection software that help doctors and patients detect
cancers in their lungs and runs on device CPU which means no mobile data will be required.
1.5 AIMS and OBJECTIVES

 Detect lung cancer stages in humans

 Offer solutions on how to treat various stages of lung cancer
 Integrate technology in the medical industry
 Provide information on how to prevent lung cancer to the coming generations
 To increase doctors acceptance of latest technology
 To reach 100% accuracy in our AI model

1.6 METHODS AND INSTRUMENTS

To implement such a project, we need to make use at the open-source technologies to help us
have a functional AI model.

Methods

Rapid Application Development

Rapid Application Development or RAD is actually an agile project management strategy which
is quite popular in the case of software development or application development.
The major advantage of this particular approach is the faster project turnaround that makes it a
much more attractive choice for the app developers that work in a fast-paced environment like
app development. This particular rapid pace is quite possible by a focus on minimizing the entire
planning stage as well as maximizing the prototype development in RAD. By means of reducing
the entire planning time as well as emphasizing the prototype iterations, it allows project
managers to measure the entire process accurately as well as communicate in real time on
different evolving issues as well as changes. This leads to greater efficiency, effective
communication and faster development.
Image Acquisition
In this phase, images of lung cancer x-rays are gathered using digital media like camera, mobile
phones etc. with desired resolution and size. The images can also be taken from web. The
formation of database of images is completely dependent on the application system developer.
The image database is responsible for better efficiency of the classifier in the last phase of the
detection system.
Image Segmentation
This phase aims at simplifying the representation of an image such that it becomes more
meaningful and easier to Analyse. As the premise of feature extraction, this phase is also the
fundamental approach of image processing.

GITHUB
A version control system that helps developers track the changes of their software projects and
also gives an online repository where you can keep your project files. GitHub makes
development faster and easier for the developer.

Instruments

1. Android Studio

2. Mobile phone & laptop

3. TensorFlow

4. Google Core Lab

5. Keras Library

6. Dataset

7.Matlab

1.7 RESULTS AND FUTURE SCOPE OF THE PROJECT

At the outset, we note that on a dataset with multiple class labels, random guessing will only
achieve an overall accuracy of 2.63% on average, but with the use of AI trained model our
application will have an accuracy of 85% +. These results with continual training and using
architectures like AlexNet and GoogleNet.
In the application of lung cancer identification, the manifestation symptoms are not obvious, so
early diagnosis is very difficult whether it is by visual observation or computer interpretation.
However, the research significance and demand of early diagnosis are greater, which is more
conducive to the prevention and control of lung cancer and prevent its spread and development.
The best image quality can be obtained on an x-ray increase the complexity of image
preprocessing and reduce the recognition effect. In addition, in the early stage of lung cancer
occurrence, even high-resolution images are difficult to analyze. It is necessary to use images
which are in gray.

The future of this project will be to have a 100% accuracy in detection of lung cancer stages and
give informed decision to doctors on how to deal with the tumour and control them so we can
have low deaths.

1.8 SIGNIFICANCE OF THE RESEARCH

This research will help in the improvement of cancer detection in patients. This research will
improve the quality of image production and also help in the advancement of cancer researches
and prevention cases. The use of such technology in the medical industry shows us how deep AI
and technology is impacting the world and how this technology can be manipulated to make our
lives simpler.

1.9 DELIMITATIONS AND LIMITATIONS OF THE RESEARCH

Limitations

1. Small dataset size problem

2. Noise on images
3. Lighting problems
4. Occlusion problems
5. Detection speed limits

Delimitations
 1. Data amplification and generation
2. Transfer learning and fine-tuning classical model
3. Reasonable network structure design

1.10 BUDGETAND TIMELINES

Research -120 usd

Laptop – 450 usd

Project estimated time 4 months

1.11 CONCLUSION

At this time, the early detection of cancer and precursor lesions offers a largely unmet potential
to reduce morbidity and mortality from malignancies. Improvements in screening quality control
and quality assurance can lead to gains in both sensitivity and specificity, thus improving
performance and reducing avoidable costs. New screening technologies focused not only on
solid tumors, but also on genetic and molecular markers of risk and disease, are under
development and eventually will replace the existing screening modalities. Ultimately, screening
failures will be measured not by death from cancer, but by the development of invasive disease.
Nevertheless, it must be emphasized that there is an unmet potential of a greater benefit from
cancer screening that is achievable today. As we anticipate these exciting new developments, we
must be reminded of the value of the technologies at hand, and their current underutilization.
There is great potential within our reach if our health care system will dedicate itself to achieving
an organized and systematic approach to early cancer detection.