Leukemia Detection Through WBC's Microscopic Blood Images Using Deep Learning Algorithm

Leukemia Detection Through WBC’s
Microscopic Blood Images Using Deep

Learning Algorithm
Syeda Munam Ali Jaffary BM-16010
Javaria Fatima BM-16022
Nisma Asif BM-16029
Ramsha Hafeez BM-16032
Supervised by
Madam Rehana Kouser
Co-Supervised by
Mr. Mustafain Ali
Bachelor of Engineering in Biomedical Engineering

Department of Biomedical Engineering
NED University of Engineering & Technology
Karachi, Pakistan
October 9, 2020
2
DECLARATION
We hereby declare that this project report is based on or original work except for citations and
quotations which have been duly acknowledged. We further declare that this report has not been
previously and concurrently submitted for any other degree or to other institutions.
Names Roll Numbers Signature

_____________________ _____________________ _____________________
_____________________ _____________________ _____________________
_____________________ _____________________ _____________________
_____________________ _____________________ _____________________
3
CONSENT LETTER
The contributors of project titled:
“_______________________________________________________________________________________________________________
____________________________________________________________________________________________”
on this ____ day of _________________, hereby consent that outcome of our Final Year Design Project (FYDP), in any form (hardware,
software, articles, patent, award etc.) shall be sole property of NED University of Engineering and Technology and must be submitted to
department prior to graduation.
Furthermore,
❖ Supervisor and co-supervisor shall suggest the order of name(s) of all FYDP members in any achievement.
❖ The FYDP members must not submit/produce/publish work related to FYDP anywhere without written consent from supervisor and co-
supervisor.
❖ The supervisor/co-supervisor/advisor shall not produce any kind of work from FYDP report without including names of all FYDP
members.
❖ The NED University of Engineering and Technology/department/relevant funding bodies must always be properly acknowledged in the
work produced.
❖ The supervisor/co-supervisor/advisor and FYDP members reserve the right to reproduce and distribute the FYDP report for personal or
professional use only.
❖ In case of potential intellectual property from FYDP; the supervisor and co-supervisor will inform the concerned chairman to restrict
the FYDP report submission to departmental library.
❖ In case of any commercial activity, visit, project display; FYDP members need to obtain a prior written permission from concerned
chairman forwarded by supervisor/co-supervisor.
❖ Failing to adhere with guidelines; the NED University of Engineering and Technology reserves the right to take any action against
FYDP member(s).
❖ It will be the responsibility of all the FYDP members to participate at all stages if applied and selected for a seminar/ competition/
conference.
Details of FYDP members
S.No. Full Name (s/o) (d/o) Father’s Name Seat No. Enrollment No. Signature
1.
2.
3.
4.
Details of Supervisor/Co-Supervisor/Advisor
Supervisor Co-Supervisor Advisor (if any)
Name,
Designation,
Affiliation.
Signature
Endorsement from CBM, NED University of Engineering and Technology

4
APPROVAL FOR SUBMISSION
I certify that this project report entitled LEUKEMIA DETECTION THROUGH WBC’S
MICROSCOPIC BLOOD IMAGES USING YOLOV3 DEEP LEARNING ALGORITHM was
prepared by SYEDA MUNAM ALI JAFFARY, JAVARIA FATIMA, NISMA ASIF and
RAMSHA HAFEEZ have met the required standard for submission in partial fulfilment of the
requirements for the award of BE in Biomedical Engineering at NED University of Engineering and
Technology, Karachi.
Approved by Supervisor
Signature: _________________________
Name: Madam Rehana Kouser
Assistant Professor
NED University of Engineering and Technology.
Approved by Head of Department
Signature: _________________________
Professor Dr. Ali Raza Jafri
Chairman
NED University of Engineering and Technology.

5
ACKNOWLEDGEMENT
We would like to acknowledge the efforts and thank everyone who had contributed to the
successful completion of this project. We would like to thank our families and our teachers for
their support and consideration towards us throughout our educational journey. Also we would
like to express our gratitude towards our research supervisor, Madam Rehana Kouser and our
co-supervisor, Mr. Mustufain Ali for their invaluable advice, guidance and their enormous
patience throughout the development of the project.
6
LEUKEMIA DETECTION THROUGH WBC’S

MICROSCOPIC BLOOD IMAGES USING DEEP LEARNING
ALGORITHM
ABSTRACT
The new era has brought an advent of Artificial Intelligence application in the healthcare
sector, which has helped revolutionize it on a global scale. Tasks that took months are now
being completed within a span of days and maybe less, the complex tasks are not so complex
anymore, tasks requiring only trained professionals does not require supervision of highly
trained personal. Artificial intelligence is used for different distinct purposes in healthcare like
the medical diagnostics, patient record keeping, designing hospitals, or running a virtual
diagnostic. Our project uses the AI‘s deep learning algorithm in order to build a leukemia
diagnosis model that will reduce the diagnosing time of the disease to give a better chance at
survival and will be more reliable, accurate, precise and cost effective. It will be a better
solution to the existing traditional testing methods which takes 1 to 3 weeks to confirm a
leukemia diagnosis. This is important because of the increasing leukemia cases. The number
of patients is increasing with every passing minute globally and so is the load on the diagnosis
laboratories, the traditional testing alone cannot bear that much load on its own, a new and
effective disease detection is needed.
7
TABLE OF CONTENTS
DECLARATION 2
CONSENT LETTER 3
APPROVAL FOR SUBMISSION 4
ACKNOWLEDGEMENT 5
ABSTRACT 6
TABLE OF CONTENTS 7
LIST OF TABLES 9
LIST OF FIGURES 10
LIST OF SYMBOLS/ABBREVIATIONS 12
LIST OF APPENDICES 13
CHAPTERS
1. INTRODUCTION 14
1.1 Problem statement 16
1.2 Objective 16
2. LITERATURE REVIEW 17
2.1 White blood cells-leukocytes 17
2.1.1 WBCs Morphology 18
2.1.1.1. Normal WBCs 18
2.1.1.2. Abnormal WBCs 18

8
2.2 Leukemia 20
2.2.1 Types of leukemia 20
2.2.2 Leukemia recorded cases 22
2.3 Traditional leukemia testing 23
2.4 Artificial intelligence 24
2.5 YOLO – You Only Look Once 26
2.5.1 How does it work? 27
2.5.2 Implementation in detection 28
3. MATERIAL AND METHODS 31
4. RESULTS 41
5. DISCUSSION 45
6. CONCLUSION 46
REFERENCES 48
APPENDICES 52
APPENDIX I: Plagiarism Report 52
APPENDIX II: Log of meetings with supervisor 53

9
LIST OF TABLES
TABLE TITLE PAGE
Table 1.1: Reveals number of recorded leukemia cases based on different age groups in
Pakistan 15
Table 2.1: Represent the total identified leukemia individuals in Yemen during the span of
2010-2014. 22
Table 2.2: Documented cases of leukemia based on varying affected age groups in Yemen.
23
10
LIST OF FIGURES
FIGURE TITLE PAGE
Figure 1.1. Recorded leukemia cases globally by WHO 14
Figure 1.2. Leukemia, its types and the people affected by it in northern areas of Pakistan by
Bangladesh journal of Medicine science Vol.18 no.02 April'09 ; original article: prevelance of
acute and chronic forms of leukemia in various regions of khyber phakhtunkhwa, pakistan by
shujaat ahmad 15
Figure 2.1.1. Types of White blood cells by Blood Cells | Biology for Majors II. 17
Figure 2.1.2. Leukocytes. (Micrographs provided by the Regents of University of Michigan

Medical School © 2012) 18
Figure 2.1.3. Toxic granulation by White Blood Cell Cases Flashcards | Memorang. 18
Figure 2.1.4. Toxic vacuolization by White Blood Cell Cases Flashcards |Memorang. 19
Figure 2.1.5. Dohle bodies by Renata, “My Notes for USMLE — REACTIVE CHANGES IN
PMN Döhle bodies (RER...” [Online] 19
Figure 2.1.6. Hyper segmented neutrophil by imagebank | hematology| “Hypersegmented

neutrophil.” [Online]. 19
Figure 2.2.1. Types of leukemia by lexymaine, “Acute Myeloid Leukemia - Know Your
Health Series #2 — Steemit.” 2018 21
Figure 2.2.2. Identified patients with different types of leukemia 22
Figure 2.2.3. Documented cases of leukemia based on the varying ages of the patients 23
Figure 2.4.1. Trend of AI by Asavari Patil, “Artificial Intelligence Market Size, Share | AI
Industry Analysis - 2025,” Jul. 2108. 24
Figure 2.4.2. Increase use of software in solving problems by “Artificial Intelligence Market
Size, Share | AI Industry Analysis - 2025,” Jul. 2108. 25
11
Figure 2.4.3. Increasing use of AI in Pakistan by Z. Hoodbhoy, B. Hasan, and K. Siddiqui,

“Does artificial intelligence have any role in healthcare in low resource settings?,” J. Med.
Artif. Intell., vol. 2, no. 0, pp. 13–13, Jul. 2019, 25
Figure 2.4.4. Increasing leukemia market due to increase in leukemia cases by “Leukemia
Therapeutics Market Global Forecast - 2024 | marketsandmarkets.” 26
Figure 2.5.1. Subset of AI by S. Singh, “Cousins of Artificial Intelligence | by Seema Singh |

Towards Data Science 26
Figure 2.5.2. Working basics of YOLO by “YOLO object detection with opencv -
pyimagesearch,” Adrian Rosebrock, 2018. 27
Figure 2.5.3. The working basics of our leukemia detector 28
Figure 3.1. Annotation for leukemia 34
Figure 3.2. Another annotation for leukemia 34
Figure 3.3. Annotation for no leukemia 34
Figure 3.4. Another annotation for no leukemia 35
Figure 3.5. Label files 36
Figure 3.6. Obj.name file 37
Figure 3.7. Obj.data file 37
Figure 3.8. Training file outlook 38
Figure 3.9. Methods used to train the model 40
Figure 4.1. First leukemia detection 41
Figure 4.2. Another detected leukemia 42
Figure 4.3. No leukemia detected 43
Figure 4.4. Another no leukemia detection 43
Figure 4.5. Error rate 44

12
LIST OF SYMBOLS / ABBREVIATIONS
WBC White Blood Cell
AML Acute Myelogenous Leukemia
ALL Acute Lymphocytic Leukemia
CML Chronic Myelogenous Leukemia
CLL Chronic Lymphocytic Leukemia
AI Artificial Intelligence
YOLO you only look once
k-NN K-Nearest Neighbor
SVM Support Vector Machines

13
LIST OF APPENDICES
APPENDIX TITLE PAGE
APPENDIX I: Plagiarism Report 53
APPENDIX II: Log of meetings with supervisor 54

14
CHAPTER 1
INTRODUCTION
The White blood cells (WBC) have a vital part in the diagnosis of different blood diseases and
disorders. The variations in the WBCs morphology is the identifying factor for the detection
and diagnostics of variety of diseases; Leukemia being one of the diseases. Leukemia is one of
the commonly prevalent blood diseases. Figure 1.1 represents the recorded leukemia cases
throughout the world.
Figure 1.1. Recorded leukemia cases globally by World Health Organization
There are four types of leukemia: acute myelogenous leukemia (AML), acute lymphocytic
leukemia (ALL), chronic myelogenous leukemia (CML) and chronic lymphocytic leukemia
(CLL). Their pervasiveness in an individual varies on the basis of the age and the environment
they are exposed to. As already mentioned, there are different types of leukemia, some of which
are common in the children than the adult while some are more prevalent in older population
than the children. Figure 1.2. shows Leukemia, its types and the people affected by it in
northern areas of Pakistan.
15
Figure 1.2.Leukemia, its types and the people affected by it in northern areas of Pakistan by
Bangladesh journal of Medicine science Vol.18 no.02 April'09 ; original article: prevelance of acute
and chronic forms of leukemia in various regions of khyber phakhtunkhwa, pakistan by shujaat
ahmad,.
Table 1.1 reveals number of recorded leukemia cases based on different age groups in Pakistan.
The recognition of leukemia is becoming easier and easier with each passing day. Leukemia is
among those diseases where early detection is important as it gives higher survival chances to
the patient, the earlier it is detected the faster the treatment can begin and therefore, better
survival chances for the individual. There has been an increase in the identified cases which
means on time treatment for those identified patient which, therefore, may lead to an increase
in the chances of their survival.
16
Problem statement
Traditionally, leukemia can be spotted by running the normal blood analysis, a bone marrow
testing along with the complete cell count and running a comparison between the number of
RBC’s and WBC’s. This conventional technique for counting blood cells with a microscope
manually takes a lot of time and energy as well as it is one of the most costly procedure in
clinical hematology laboratory. With the upsurge in the quantity of cases, it becomes essential
to build a quicker, easier to decipher, trustworthy and precise testing method. Therefore, for
surmounting this financial and time barrier as well as for running an efficient and reliable
leukemia diagnostic using WBC’s microscopic images; AI deep learning algorithm, YOLO is
used. Following the trail of the morphological changes in WBCs is a more reliable and accurate
way of identifying leukemia and this does exactly that.
Objective
The objective of this project is to research if the application of the artificial intelligence (AI)
deep learning algorithm; YOLO for the detection of any changes in WBCs morphology would
result in the timely, reliable and accurate diagnosis of leukemia. The expectation is that it
would yield in a result that would be an accurate, reliable and fast detection of leukemia.
17
CHAPTER 2
LITERATURE REVIEW
2.1. White blood cells- leukocyte

Both the RBCs and the WBCs originate from the hematopoietic stem cells which are primarily
found in the bone marrow and together they constitute for approximately 45% of human blood,
while the rest is simply composed of the plasma. Taking a few peeks into their morphology
would give very valuable and crucial details for the recognition of numerous blood borne
disorders and diseases. But for the sake of this project our focus is only on the White blood
cells.
White blood cells are an essential element of the immune system which offers human body,
the defense it needs against any infections and the alien bodies.
White blood cells are classified into neutrophil, eosinophil, basophil, monocyte and
lymphocyte, with each of them consisting of a nucleus of their own. These are grouped into
different classifications, two of the which are the most common categories, based on the
presence or absence of cytoplasm- granulocytes or agranulocytes, respectively - and myeloid
or lymphoid cell lineage- granulocytes or lymphocytes, respectively.
Figure 2.1.1. Types of White blood cells by Blood Cells | Biology for Majors II.
Each of these different types of WBCs are responsible for a different function within a human
body. For instance, granulocytes play a key role in immune response to fungi, bacteria and any
parasites while the Lymphocytes are highly specialized antigen which are able to target a
specific infective agent.
18
Each WBC is essential and examining morphology of each of them offers an insight in health
of an individual. Running an analysis on the changes in morphology helps with the detection
of diseases and disorders.
2.1.1. WBC Morphology
2.1.1.1. Normal WBCs
Figure 2.1.2. Leukocytes. (Micrographs provided by the Regents of University of Michigan Medical
School © 2012)
2.1.1.2. Abnormal WBC
Some of the anomalies observed in WBCs are as follows:
o Toxic granulation
Uniformly dispersed grains; large or medium,

throughout the cytoplasm of segmented
polymorphonuclear neutrophil leukocytes.
They can be noticed in individuals suffering
from serious bacterial infection, toxemia of
pregnancy, vasculitis, or a patient going
through chemotherapy.
Figure 2.1.3. toxic granulation by White Blood
Cell Cases (Heme-Onc) Flashcards | Memorang.
19
o Toxic vacuolization
The spotless clear round areas aimlessly

scattered throughout the neutrophils’
cytoplasm. These can be noticed in
individuals suffering from serious
infection.
Figure 2.1.4. toxic vacuolization by White Blood

Cell Cases (Heme-Onc) Flashcards | Memorang
o Dohle bodies
Typically varying from 1 to 3 µm in

thickness are blue, oval bodies
originating in RNS. Depending on the
kind and gravity of the diseases or
disorders there can be either single or
numerous dohle forms. They can be
noticed in individuals experiencing
serious infections, or disorders like: Figure .2.1.5. Dohle bodies by Renata, “My Notes
for USMLE — REACTIVE CHANGES IN PMN Döhle
May-Hegglin anomaly and the Chediac
bodies (RER...” [Online]
Higashi disorder or in patients receiving
chemotoxic medications.
o Hyper segmented neutrophil
A neutrophil comprising of more than

five lobes. These can be seen in
individuals with the vitamin B12
deficiency. Hypersegmented neutrophil
is a symptom of a nuclear maturing
defect. Figure 2.1.6. hyper segmented neutrophil by
imagebank |hematology | “Hypersegmented
neutrophil.” [Online].
20
2.2. Leukemia
Leukemia suggests the cancer of blood cells. Blood cells are divided into three major categories
including, Platelets, White Blood Cells (WBCs) and the Red blood cells (RBCs). Leukemia
refers to the cancer or the malignancy of the White blood cells.
As already established in the previous section, WBCs have a crucial role in the development
of immune system as it defends the body from the viruses, fungi, and the bacteria, as well as
from unusual cells and other unknown substances. In leukemia, the WBCs are incapable of
carrying out their roles naturally. They lose their genuine count, size and shape.
2.2.1. Types of leukemia
Leukemia can be categorized based on the inception of acute or chronic. In acute leukemia,
cancer cells multiply quickly. In chronic leukemia, cancer cells multiply slowly and early
indications may be very mild.
Leukemia is further classed based on the type of cells. Leukemia involves two types of cells
myeloid cells and lymphocytes; involvement of myeloid cells implies to the myelogenous
leukemia and contribution of lymphocytes implies to the lymphocytic leukemia.
Based on type of cells, there are four types of leukemia:
o Acute myelogenous leukemia (AML):
Acute Myelogenous Leukemia can happen in children and matured grown individuals. As
per the Surveillance, Epidemiology, and End Results Program of the National Cancer
Institute (NCI), around 21,000 new instances of AML are analyzed every year in the United
States. This is the most well-known kind of leukemia. The five-year endurance rate for
AML is 26.9 percent.
21
o Acute lymphocytic leukemia (ALL):
Ac lymphocytic leukemia (ALL) occurs normally in children. The NCI estimates around
6,000 new instances of ALL are analyzed every year. The five-year endurance rate for ALL
is 68.2 percent.
o Chronic myelogenous leukemia (CML)
Chronic myelogenous leukemia (CML) mostly affects the adults. Around 9,000 new
instances of CML are analyzed every year. As indicated by the NCI, the five-year
endurance rate for CML is 66.9 percent.
o Chronic lymphocytic leukemia (CLL)
Chronic lymphocytic leukemia (CLL) is destined to influence individuals above the age of
55. It's once in a while found in youngsters. As indicated by the NCI, around 20,000 new
cases of CLL are analyzed yearly. The five-year endurance rate for CLL is 83.2 percent.
Furry cell leukemia is an extremely unusual subtype of CLL. It gets its name from the
presence of the malignant lymphocytes under an amplifying lens.
Figure 2.2.1. types of leukemia by lexymaine, “Acute Myeloid Leukemia - Know Your Health
Series #2 — Steemit.” 2018
22
2.2.2. Leukemia recorded cases
There is a rise in the amount of recorded cases of leukemia. Typically, the testing for its
diagnostics takes from days to a week and a trained interpreter for interpretation of the results.
With the continuous rise in the recorded leukemia cases, it becomes essential to develop a
faster, easier to interpret, reliable and accurate testing method.
Table 2.1 represent the total identified leukemia individuals in Yemen during the span of 2010-
2014.
years ALL AML CLL CML

2010 69 74 35 44
2011 92 69 27 58
2012 52 61 47 42
2013 129 82 60 94
2014 169 117 81 120
The table can be inferred into the graph that evidently reveals in the increase of the documented
leukemia cases over a four-year span.
Registered cases of Leukemia ALL
AML
180
CLL
160
140
number of cases
CML
120
100
80
60
40
20
0
2010 2011 2012 2013 2014
years
Figure 2.2.2 identified patients with different types of leukemia
23
Occurrence of different kinds of Leukemia is also affected by age. Table 2.2 trend coincides
with this statement.
Table 2.2 represents the documented cases of leukemia based on varying affected age groups
in Yemen.
Age ALL AML CLL CML

1-10 292 59 0 3
11-20 138 76 0 11
21-30 49 105 0 52
31-45 15 99 26 102
46-60 9 39 105 108
>60 8 25 119 82
Documented cases of leukemia based on varying

ages
350 ALL
300 AML
Recorded cases
250
CLL
200
CML
150
100
50
0
1-10 11-20 21-30 31-45 46-60 >60
Age
Figure 2.2.3. documented cases of leukemia based on the varying ages

of the patients
2.3. Traditional leukemia testing
The traditional leukemia testing is using the normal blood work and bone marrow testing; the
latter is invasive, painful and the result takes one to several weeks to reach to a diagnosis.
24
2.4. Artificial intelligence:
Globally, Artificial intelligence is in the process of claiming a share in various fields of

application, in the upcoming decade and Pakistan is no exception to this. The application of
AI is increasing in all sectors especially in the health care as its end results are more reliable
and accurate as well it help people focus their attention on things that can’t be solved using
computation power. Its use is increasing day by day and it is forecasted to do so more rapidly
in the near future,
Figure 2.4.1. trend of AI by Asavari Patil, “Artificial

Intelligence Market Size, Share | AI Industry Analysis -
2025,” Jul. 2108.
The increase in the use of software in the future is the leading result of the increased use of the
Machine learning algorithms as a solution to the different problems faced in this era. The bar graph
(taken from an online source) backs up the fact that the use of software is on the continual rise and
it has more market potential than the rest.
25
Figure 2.4.2. increase use of software in solving problems by “Artificial

Intelligence Market Size, Share | AI Industry Analysis - 2025,” Jul. 2108.
According to an article based on a recent survey at a prestige hospital, the use of AI in different
healthcare sectors is increasing in Pakistan and is predicted to behave in a similar way in the future.
The following chart was the result of that survey.
Figure 2.4.3. increasing use of AI in Pakistan by

Z. Hoodbhoy, B. Hasan, and K. Siddiqui, “Does
artificial intelligence have any role in healthcare
in low resource settings?,” J. Med. Artif. Intell.,
vol. 2, no. 0, pp. 13–13, Jul. 2019,
Our project is specifically targeting the diagnosing and detection of leukemia using deep
learning. The market for the leukemia diagnostic has an increasing trend and claiming some of
26
it through unconventional methods is very much possible as traditional methods cannot meet
the increasing demand on its own. The bar chart below clearly shows the increase in the
specified market over a defined period of time and for different regions around the world.
Figure 2.4.4. increasing leukemia market due to increase in

leukemia cases by “Leukemia Therapeutics Market Global
Forecast - 2024 | MarketsandMarkets.”
2.5. YOLO – You Only Look Once:
A smart convolutional neural network (CNN) that is used for doing the object detection – You
Only Look Once. Neural networks are a part of the deep learning that themselves are part of the
machine learning which is a subset of the AI.
Figure 2.5.1. subset of AI by S. Singh, “Cousins of

Artificial Intelligence | by Seema Singh | Towards
Data Science
27
2.5.1. How does it work?

The YOLO algorithm applies a single neural network to the full image; in simpler words, it
only looks at an image once, then divide up the image into areas and the forecasts bounding
boxes and probabilities for each of the divided region. The bounding boxes are assigned some
weights by the projected probabilities.
Figure 2.5.2 working basics of YOLO by “YOLO object detection with OpenCV -
PyImageSearch,” Adrian Rosebrock, 2018.
YOLO, with only a single CNN can simultaneously predicts several bounding boxes with their
probabilities. YOLO model has a lot of benefits over any of the other object detection
techniques, procedures or methods used; like it is really fast, its time efficient or that it looks
at the entire image when training or testing and subtly encodes contextual information about
classes as well as their appearance.
28
The following chart is representation of how it worked for our project.
Figure 2.5.3. The working basics of our leukemia detector
2.5.2. Implementation in detection:
The researchers’ working in the field of diagnostics are more and more inclined to use the
neural networks for running the diagnostics. The articles that we are referencing to ran different
types of deep learning algorithms to run the leukemia diagnostics. They are finding this method
much more reliable and accurate than the other methods being applied in the healthcare for the
same purposes. Not only is this method, regardless of the type of neural network applied, is
more reliable, precise, and accurate but it is time efficient and effective than the rest exhaustive
methods being applied. Some articles that we related to have run detection using only one
Machine learning algorithm each while some ran the analysis with multiple and compared their
results, we are applying only one algorithm like the most.
The techniques that were compared are as follows along with their brief description. The
algorithms and techniques taken under observation takes the input of the color of the images
of the stained blood smears, preprocesses them, performs image segmentation, feature
extraction and also classifies them to see whether the patient is affected by leukemia, all of
these are done on White blood cells microscopic images.
29
o Support Vector Machines (SVM)

A binary classification algorithm which is used to classify sample blood images to myeloid
stem cells and lymphoid stem cells which mark them as Acute Lymphoblastic Leukemia
and Acute Myeloid Leukemia. The number of nuclei lobes, ratio of nuclei to cell, ratio of
perimeter to nuclei and entropy are the features it focuses on. Using Support Vector
Machines (SVM), the achieved accuracy was approximately 92%.
o K-Nearest Neighbor (k-NN)
It classifies blasts in leukemic cells and classified the cells into Acute Myelogenous
Leukemia (AML) and Acute Lymphocytic Leukemia (ALL) with an accuracy of 80%. The
main features that were extracted from leukemic blood images were to represent size, color
and shape and k-values and distance metric were tested many times. This algorithm
classifies new objects based on similarity measures under the assumption that similar
things exist in close proximity. When using the value of k=4 and cosine distance metric,
the k-NN classifier gave good results.
o Naïve Bayes
It only small set of data is needed for the training and it can estimate several parameters
vital for the classification. The posterior probability can be found using Naïve Bayes
Classifier using Bayes rule. Being a controlled learning classifier, its parameter evaluation
is established on the maximum likelihood scheme. They are believed to be the independent
variables and saved, after which the probability of prior p(c) was counted.
o Neural Networks- Deep Learning algorithm
Neural networks are used to classify blood smear images into regular blood cells and
leukemic blood cells to diagnose leukemia. The preprocessing steps include the image
preprocessing, image clustering and image segmentation using the Principal Component
Analysis (PCA) is achieved by extracting the principal components which become the
input to the Neural Network classifier whose purpose is to classify between normal and
abnormal- leukemia affected- cells.
Then again, another deep learning algorithm was analyzed where the blood cell images of
both patients having leukemia history and normal people were taken as the dataset. The
dataset was then divided among the training and testing data. The input layer defines the
30
image size to the CNN which communicates to the height, width and number of channels
of the image. The convolutional layer or the second layer consists of the neurons that
creates the connection between the sub region of the image and the layers that are taken as
the output before it. The convolutional layer trains and accustoms itself to the features
localized by these regions after scanning the image. It has the most accuracy amongst all
the other algorithms.
Our review leads us to believe that YOLO was the most fast, effective, and efficient way of
achieving our target, which it was. Going through the research paper we learned that each
algorithms like everything else had its own pros and cons but the one thing common with the
other algorithms was the fact that they were not as easy to handle and fast as YOLO. Since
YOLO only runs analysis on the image once it trains faster than the most and it had the error
rate less than the other algorithms.
Every hidden layer used within the neural networks has a purpose, they assign weights based
on the characteristics that they are learning, for instance the layer defining the color and
boundary both are different, one requires more preciseness then the other so both are part of
different layers, YOLO is better at assigning these layers and finding the characteristics and is
much than the rest at doing so. This is particularly one reason that so many individuals are
inclined to run their detections using this specific algorithm even if their area of detection
varies. Other than that YOLO is able run detection with multiple classes simultaneous, this
gives it another edge over others which are limited to running one or at most a handful classes
simultaneously.
The articles we read included custom detection using YOLO on random objects as well as for
the diagnostics. The researchers who used it for detecting leukemia were satisfied with the
results, just like expected their detection ran with accuracy and generated reliable results;
overall reducing the diagnosing time for leukemia especially with the comparison of the
traditional testing and diagnostic method of the disease.
We opted for YOLO after coming to conclusion that most of the article ran their diagnostics
with either YOLO alone or ran a comparison with YOLO where they discovered that object
detection is much more accurate with it. And since our objective was to find a reliable, fast,
precise and accurate detection method, YOLO was our algorithm of choice.
31
CHAPTER-3
MATERIAL AND METHODS
There were a few requirements we needed for the project and they are as follows:
Dependencies:
• CUDA
• OpenCV
• Image annotation tool
Files:
• Labelled custom dataset

• Configuration (.cfg) file
• Obj.names file
• Obj.data file
• Train.txt file
• Backup folder
We are building our model on Cloud using the Google Colab which allows us to use all the
necessary dependencies online such as OpenCV and CUDA. We need these two dependencies
to install darknet to build our model. OpenCV is necessary so that a wide range of variety of
image types are supported and CUDA is important to increase the computation power -using
the GPU- to increase the efficiency of the computation; since it requires hours on end to train
a model. Having GPU increases the efficiency and accuracy of the trained model.
We have built our model on the existing algorithm that we obtained from the opensource github
repository, but we have tailored it upon our needs. The steps that we took are as follows along
with their respective codes that we used:
• The Cloning the darknet AlexeyAB repository. This is the base of the project.
! git clone https://github.com/AlexeyAB/darknet

32
• Then making a few required adjustments to the Makefile in the darknet repository.
These changes were required for the OPENCV, GPU and CUDNN; all of these were
needed to be made 1 to enable them; so that the darknet will have access to these
dependencies.
%cd darknet
! sed -i 's/OPENCV=0/OPENCV=1/' Makefile
! sed -i 's/GPU=0/GPU=1/' Makefile
! sed -i 's/CUDNN=0/CUDNN=1/' Makefile
• Building the darknet by use of the make command. This ‘make’ command will run all
the necessary executions to build the darknet directory.
! make
• When making a custom detector we use pretrained weights so that it forms a basis on
which upon further training the weights are added on and accurate and precise layers
bias is generated from. If we use random weights that we have roughly taken on from
a guess as a basis the model, it will take additional hours to train and still it wouldn’t
be as precise as needed. In conclusion taking rough estimates for weights hinders the
productivity and efficiency of the model. Using pretrained weight is the better option.
So, we used pretrained weights for our model.
! wget https://pjreddie.com/media/files/yolov3.weights
• Along with the preset weights we require a helper function. This function is what will
create the bounding boxes around the detected images:
def imShow(path):
import cv2
import matplotlib.pyplot as plt
image = cv2.imread(path)
height, width = image.shape[:2]
33
resized_image = cv2.resize(image,(3*width, 3*height), interpolation = cv2.INTER_

CUBIC)
fig = plt.gcf()
fig.set_size_inches(18, 10)
plt.axis("off")
plt.imshow(cv2.cvtColor(resized_image, cv2.COLOR_BGR2RGB))
plt.show()
def upload():
from google.colab import files
uploaded = files.upload()
for name, data in uploaded.items():
with open(name, 'wb') as f:
f.write(data)
print ('saved file', name)
def download(path):
from google.colab import files
files.download(path)
• Since we are building a custom detected we had to create our own dataset which
contained annotation for approximately 2500 images. These custom annotated images
were the WBC microscopic images that we are working on for the detection of
leukemia. The dataset contained image sets for leukemia and non-leukemia WBCs.
The annotations for the images are the labels over the objects that we need to be
detected. For our case our images needed annotations for the labels no leukemia and
leukemia over the WBCs since our objective is to detect leukemia on the basis of the
WBC morphology.
These annotations were made using an online annotation tool.
34
The annotation tools are used to label the detected region with the accuracy.
For example, the leukemia detection:
Figure 3.1.annotation for leukemia
Figure 3.2. another annotation for

leukemia
And for no leukemia detection
Figure 3.3. annotation for no leukemia

:
35
Figure 3.4. another annotation for no

leukemia
The purpose of annotation is that it will give the coordinates of the region that we want
the model to train itself for the detection.
When training any detector, it requires the coordinates of the object that it is to learn
and the format of each of these coordinates is different. The format that yolo needs is
as follows:
[Label number in order x y width height]
The label files will have the number in the form of array like that.
For leukemia the label will look like this:
For no leukemia the label will look like this:
Correct labelling is of utmost importance. While labelling maintaining the order of the
labels is very important because the algorithm will read the numbers and learn from it.
It will connect the label files to the object files- files containing the label names- it will
read the first column and connect it with the order of class name mentioned in the other
file. Since it reads labels like a matrix format.
36
the dataset contains the images as well as their label files along with it; so, the file will
look like this:
Figure 3.5. label files
• After the dataset there are a set of files that are required for our model to base upon.
The first of which is the configuration file. The basic format of the configuration file
for any model remains same though there are some changes that has to be made in this
file for each model that is to be build and a number of these changes are required in
this file before and after the training.
There are few changes that the file needs before the training are as follows, these
changes were made manually because files required changes these steps several times
in the file, to be sure about the reliability of the changes, it was done manually:
o Batch and subdivision to be changed to 64 and 16 respectively.
o The number of classes to be changed to 2; we are using two classifications

in our model that are no leukemia and leukemia.
o The max_batches to be changed to 4000; this is calculated by the formula:
max_batches = 2000 * number of classes.
o The steps to 3200 and 3600; the formula it is calculated from is:
Steps = 80% of the max_batches , 90% of the max_batches
o The filters to 21 and the formula for calculation of filters is used:

Filters = (number of classes + 5) * 3
37
The Changes the configuration file needs after the training is only in its batch and
sub_division which are both to be made 1. This we did with the help of a code:
%cd cfg
!sed -i 's/batch=64/batch=1/' yolov3_custom.cfg
!sed -i 's/subdivisions=16/subdivisions=1/' yolov3_custom.cfg
%cd .
• Next, we required an Obj.names file which is simply a file consisting of the names of
classes in the order that we have annotated our images in.
For us our ordered sequence for the classes were:
1. Leukemia
2. No leukemia
This is what the file looks like:
Figure 3.6. obj.names
• The Obj.data file is also required, and this file contains the path of all the necessary
files.
Since we are working on cloud, we set the path in accordance with it.
Figure 3.7. obj.data file

38
• One of the most important files is the train.txt file, since this image contains the dataset
path that is required to train the model. This file was obtained by a python code which
converted the path of files in the dataset into one file that the algorithm needs to run
the training on.
The following python code was used:
import os
image_files = []
os.chdir(os.path.join("data", "obj"))
for filename in os.listdir(os.getcwd()):
if filename.endswith(".jpg"):
image_files.append("data/obj/" + filename)
os.chdir("..")
with open("train.txt", "w") as outfile:

for image in image_files:
outfile.write(image)
outfile.write("\n")
outfile.close()
os.chdir("..")
The training file will look like this:
Figure 3.8. training file outlook

39
• The backup folder is also required to save the changes in the weight file after every few
hundred iterations- the algorithm runs thousands of iterations during its training- these
files are necessary to backup otherwise each time the training is interrupted for any
reason the training will have to start from scratch and all the progress will be lost.
• After these files were made and linked to the algorithm the trained phase started which
lasted for a few days.
For training the following code is used:
For training first time without using weights from backup:
!./darknet detector train data/obj.data cfg/yolov3_custom.cfg darknet53.conv.74 -

dont_show
For training as many times after the first using weights from backup. It continues where
the last saved training model was left at.
!./darknet detector train data/obj.data cfg/yolov3_custom.cfg /mydrive/yolov3/backup

/yolov3_custom_last.weights -dont_show
• Once the model was trained, the detector was ready to be tested. For testing the
following code was used:
!./darknet detector test data/obj.data cfg/yolov3_custom.cfg /mydrive/yolov3/backup/

yolov3_custom_last.weights /mydrive/images/BloodImage_00104.jpg -don’t show
imShow('predictions.jpg')
40
The following chart summarizes the methods that we applied:
Figure 3.9. methods used to train the model

41
CHAPTER 4
RESULTS
The images were correctly detected. The WBCs that had leukemia were detected with a good
accuracy as well as the WBCs without leukemia. Following are the few of some of the test
images that were run on the custom leukemia detector.
The following pictures were detected with leukemia and they showed the WBCs affected by
leukemia, so the detection was precise.
Detected with leukemia:
Figure 4.1. first leukemia detection

42
Figure 4.2. another detected leukemia
The following images were detected to have no leukemia and the images were infact of the
WBCs that have not been affected by leukemia, so once again there was detection that was
reliable and precise.
43
Detected without leukemia:
Figure 4.3. no leukemia detected
Figure 4.4. another no leukemia detection

44
With an error rate was as low as 0.56:
Figure 4.5. error rate

45
CHAPTER 5
DISCUSSION
Normally it takes days and weeks for diagnosing leukemia using the traditional testing method
but using the leukemia detector it will takes less than an hour to detect the leukemia using the
white blood cell microscopic blood images. Not only reducing time but it will also eliminate
the need of a highly trained professional to interpret the results every time the diagnosis is run.
The articles that we are referring to use the AI Machine learning algorithms to reach to their
desired results. They ran several Deep learning algorithms for finding a fast, simple, effective,
reliable and accurate methods for the detection purposes. They all came to conclusion that the
detection using Machine Learning algorithm technique is much accurate and need of the time
that the other methods applied for the same purpose. Different algorithms were running but the
deep learning algorithms like YOLO was the most effective , accurate and time efficient
algorithm of all. The results that they obtained were with lower error rate than any other
detecting method as well as they were easier to handle and faster to implement and execute.
The results were much more reliable and accurate with the detection too. Our results were the
same way, they were in line with the results as obtained by the other researchers in their
research, the results that we obtained were with low error rate, more reliable and precise. The
results were easier to interpret too.
It is, therefore, possible to use the custom detector to create the model for the leukemia
detection. The model will be much more efficient, reliable, and effective.
46
CHAPTER 6
CONCLUSION
The number of leukemia cases are increasing day by day and it is predicted to continuously
rise in the upcoming future. The load of detection and diagnosing is also increasing with the
increase in the individuals affected by the disease. The faster leukemia is diagnosed, the faster
the treatment can begin and not only that, the higher the survival chances will get. The
traditional leukemia testing will not be able to bear the load on its own for the diagnosing.
Apart from that fact, it is not fast and as efficient as the testing needs to be. Recently the
researchers are looking into different new and innovative ways of leukemia diagnosing that
will not only reduce the diagnosing time but will also increase the reliability and the efficiency
of the results. The researchers are carrying out analyzing using different methods and
algorithms but application of AI using the Deep learning algorithms for this specific detection
purpose is gaining popularity because of its efficiency, reliability and preciseness.
AI has already improved different fields of life, the things that we first thought were impossible
or difficult to achieve are now easier and doesn’t seem as complex as they once were. AI has
made life easier in different sector of our life so what not use it further to make improvements
wherever possible, if using AI can give us benefits in the health care department,
revolutionizing it in the process so why not take our chances with it. Application of AI more
specifically deep learning algorithm is full of promises for the health care sector, researching
it further and implementing the findings into our daily life could not only improve the standards
of the health care sector and living but it could save countless lives in the process.
We too have carried out an analysis on the use of the Deep learning algorithm specifically
YOLO on to the microscopic White blood cells images for the detection of the leukemia. We
trained the model using normal White blood cell microscopic images and with the leukemia
affected White blood cell microscopic images. We ran our model using a dataset of
approximately 2500 images and trained it for hours to get the desired results. The algorithm
trained faster than any of the other counterparts as well as it was easier to handle and it resulted
in detection with the error rate of less than 0.6%. Therefore, it is safe to say that the results
obtained were not only time saving but precise, accurate and reliable and easy to interpret. All
47
we have to do after training the model is run the detection on as many images as we want. It
will only take some initial time load but then it will be good to go.
The analysis was successful since we got the desired and expected results with an increased
efficiency and reliability. Therefore, it is safe to say that application of the deep learning
algorithms in the health care sector for disease diagnosis is very helpful and has the potential
to revolutionaries the healthcare sector. Just like we were able to train a model to precisely
detect leukemia, it can be used for detection of other diseases too. The results will be accurate
with the least notable error rate and increased accuracy. The detector can be converted into
application with further research which can be used whenever and wherever needed without
any professional to load it each and every single time.
48
REFERENCES
Research Papers:
[1] “Slide preparation for WBC differential by manual microscopy,” Boule.com, no. 1, pp.
1–4, 2012, [Online]. Available: https://www.boule.com/wp-
content/uploads/2019/12/white-paper-slide-preparation-for-wbc_wpb_34068-2.pdf.
[2] Zahra Hoodbhoy, Babar Hasan, and Khan Siddiqui, “Does artificial intelligence have
any role in healthcare in low resource settings? - Hoodbhoy - Journal of Medical
Artificial Intelligence,” Jul. 02, 2019.
http://jmai.amegroups.com/article/view/5049/html (accessed Aug. 29, 2020).
[3] “Artificial Intelligence (AI) in Healthcare Market Size, and Growth 2027.”
https://www.alliedmarketresearch.com/artificial-intelligence-in-healthcare-market
(accessed Aug. 29, 2020).
[4] Asavari Patil, “Artificial Intelligence Market Size, Share | AI Industry Analysis - 2025,”
Jul. 2108. https://www.alliedmarketresearch.com/artificial-intelligence-market
(accessed Aug. 29, 2020).
[5] Kavita Joshi and Onkar Sumant, “Artificial Intelligence (AI) in Healthcare Market Size,
and Growth 2027,” Jun. 2020. https://www.alliedmarketresearch.com/artificial-
intelligence-in-healthcare-market (accessed Aug. 29, 2020).
[6] Q. Wang, S. Bi, M. Sun, Y. Wang, D. Wang, and S. Yang, “Deep learning approach to
peripheral leukocyte recognition,” PLoS One, vol. 14, no. 6, pp. 1–18, 2019, doi:
10.1371/journal.pone.0218808.
[7] H. A. Werman and C. G. Brown, “White blood cell count and differential count,”
Emerg. Med. Clin. North Am., vol. 4, no. 1, pp. 41–58, 1986.
[8] A. ELEN and M. K. TURAN, “Classifying White Blood Cells Using Machine Learning
Algorithms,” Uluslararası Muhendis. Arastirma ve Gelistirme Derg., pp. 141–152,
2019, doi: 10.29137/umagd.498372.
[9] M. M. Alam and M. T. Islam, “Machine learning approach of automatic identification

49
and counting of blood cells,” Healthc. Technol. Lett., vol. 6, no. 4, pp. 103–108, 2019,
doi: 10.1049/htl.2018.5098.
[10] R. Article et al., “of Biomedical AND Pharmaceutical sciences,” vol. 3, no. 4, pp. 156–
166, 2016.
[11] Y. Nie, P. Sommella, M. O’Nils, C. Liguori, and J. Lundgren, “Automatic detection of

melanoma with yolo deep convolutional neural networks,” 2019 7th E-Health Bioeng.
Conf. EHB 2019, no. November, 2019, doi: 10.1109/EHB47216.2019.8970033.
[12] S. Shafique and S. Tehsin, “Acute lymphoblastic leukemia detection and classification
of its subtypes using pretrained deep convolutional neural networks,” Technol. Cancer
Res. Treat., vol. 17, pp. 1–7, 2018, doi: 10.1177/1533033818802789.
[13] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-
time object detection,” Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit.,
vol. 2016-Decem, pp. 779–788, 2016, doi: 10.1109/CVPR.2016.91.
[14] R. Negi and R. Mathew, “Machine Learning Algorithms for Diagnosis of Breast
Cancer,” Lect. Notes Data Eng. Commun. Technol., vol. 31, no. 01, pp. 928–932, 2020,
doi: 10.1007/978-3-030-24643-3_109.
[15] H. T. Salah, I. N. Muhsen, M. E. Salama, T. Owaidah, and S. K. Hashmi, “Machine

learning applications in the diagnosis of leukemia: Current trends and future directions,”
Int. J. Lab. Hematol., vol. 41, no. 6, pp. 717–725, 2019, doi: 10.1111/ijlh.13089.
[16] R. Janaki, “Detection of leukemia in microscopic white blood cell images using
Gaussian feature convolutional visual recognition algorithm,” J. Crit. Rev., vol. 7, no.
3, pp. 173–180, 2020, doi: 10.31838/jcr.07.03.32.
Web Sources:
[1] A. SHARMA, “How to Use Google Colab for Deep Learning and Machine Learning.”
2020, [Online]. Available: https://www.analyticsvidhya.com/blog/2020/03/google-
colab-machine-learning-deep-learning/.
[2] “A Complete guide to Google Colab for Deep Learning.” [Online]. Available:
50
https://www.kdnuggets.com/2020/06/google-colab-deep-learning.html.
[3] P. K. Sahu, “A Guide To Build Your Own Custom Object Detector Using YoloV3 | by
Prabhat Kumar Sahu | Analytics Vidhya | Medium.” 2019, [Online]. Available:
https://medium.com/analytics-vidhya/custom-object-detection-with-yolov3-
8f72fe8ced79.
[4] A. Muehlemann, “How to train your own YOLOv3 detector from scratch | by Anton
Muehlemann | Insight.” 2019, [Online]. Available:
https://blog.insightdatascience.com/how-to-train-your-own-yolov3-detector-from-
scratch-224d10e55de2.
[5] J. Nelson, “Training a YOLOv3 Object Detection Model with a Custom Dataset | by
Joseph Nelson | Towards Data Science.” [Online]. Available:
https://towardsdatascience.com/training-a-yolov3-object-detection-model-with-a-
custom-dataset-4981fa480af0.
[6] S. Nayak, “Training YOLOv3 : Deep Learning based Custom Object Detector | Learn
OpenCV.” 2019, [Online]. Available: https://www.learnopencv.com/training-yolov3-
deep-learning-based-custom-object-detector/.
[7] S. Nayak, “Deep Learning based Object Detection using YOLOv3 with OpenCV (
Python / C++ ) | Learn OpenCV.” 2018, [Online]. Available:
https://www.learnopencv.com/deep-learning-based-object-detection-using-yolov3-
with-opencv-python-c/.
[8] M. Murugavel, “How to train YOLOv3 to detect custom objects | by Manivannan

Murugavel | Medium.” 2018, [Online]. Available:
https://medium.com/@manivannan_data/how-to-train-yolov3-to-detect-custom-
objects-ccbcafeb13d2.
[9] “YOLO: Real-Time Object Detection.” [Online]. Available:

https://pjreddie.com/darknet/yolo/.
[10] R. Khandelwal, “Implementing YOLO on a custom dataset | by Renu Khandelwal |

Towards Data Science.” 2019, [Online]. Available:
51
https://towardsdatascience.com/implementing-yolo-on-a-custom-dataset-
20101473ce53.
[11] D. Zuenko, “Preparing data before training YOLO v2 and v3. #DeepFashion dataset. |
by Denis Zuenko | Good Audience.” 2018, [Online]. Available:
https://blog.goodaudience.com/part-1-preparing-data-before-training-yolo-v2-and-v3-
deepfashion-dataset-3122cd7dd884.
[12] “What is Perceptron | Simplilearn.” [Online]. Available:

https://www.simplilearn.com/what-is-perceptron-tutorial.
[13] J. Brownlee, “How To Implement The Perceptron Algorithm From Scratch In Python.”
2016, [Online]. Available: https://machinelearningmastery.com/implement-perceptron-
algorithm-scratch-python/.
[14] “GitHub - AlexeyAB/darknet: YOLOv4 - Neural Networks for Object Detection

(Windows and Linux version of Darknet ).” [Online]. Available:
https://github.com/AlexeyAB/darknet.
[15] “YoloGenerateTrainingFile/generate_train.py at master ·

theAIGuysCode/YoloGenerateTrainingFile · GitHub.” [Online]. Available:
https://github.com/theAIGuysCode/YoloGenerateTrainingFile/blob/master/generate_tr
ain.py/.
[16] “YOLO object detection with OpenCV - PyImageSearch,” Adrian Rosebrock, 2018.
https://www.pyimagesearch.com/2018/11/12/yolo-object-detection-with-opencv/
[17] “ABNORMAL MORPHOLOGY OF BLOOD CELLS.” 2012.
[18] “Blood Cell Images | Kaggle.” [Online]. Available:

https://www.kaggle.com/paultimothymooney/blood-cells.
52
APPENDIX I: Plagiarism Report

53
APPENDIX II: Log of Meetings with supervisor
NED UNIVERSITY OF ENGINEERING & TECHNOLOGY, KARACHI

DEPARTMENT OF BIOMEDICAL ENGINEERING
FYP Group Meeting with Supervisor

Meeting Summary
Meeting attended by: SYEDA MUNAM ALI (BM-16010)

JAVARIA FATIMA (BM-16022)
NISMA ASIF (BM-16029)
RAMSHA HAFEEZ (BM-16032)
Absent from
meeting:
Date: 24TH OCT 2019
Time: 11 AM
Location: NEDUET, LEJ Campus.
Agenda: Meeting No: 1
Points of Meeting:
1. Introductory meeting with the supervisor and co-supervisor.
2. Exploration for the topic.
Next Meeting: 31ST OCT 2019 Signature: (Supervisor)
54


Meeting Summary

MR. MUSTAFAIN ALI
Absent from
meeting:
Date: 31ST OCT 2019
Time: 11 AM
Agenda: Topic discussion Meeting No: 2
Points of Meeting:
1. Topic decided having engineering and medical perspective.
Next Meeting: 7TH NOV 2019 Signature: (Supervisor)
55


Meeting Summary

Absent from
meeting:
Date: 21ST NOV 2019
Time: 11 AM
Agenda: Literature Review Meeting No: 3
Points of Meeting:
1. To study about white blood cells, their types and morphologies.
2. To write a brief introduction on the topic.
Next Meeting: 28TH NOV 2019 Signature: (Supervisor)
56


Meeting Summary

Absent from
meeting:
Date: 28TH NOV 2019
Time: 11 AM
Points of Meeting:
1. To study research papers of white blood cells segmentation.
2. To make a report on important points of research papers and submit it in next
meeting.
Next Meeting: 5TH DEC 2019 Signature: (Supervisor)
57


Meeting Summary

Absent from
meeting:
Date: 5TH DEC 2019
Time: 11 AM
Points of Meeting:
1. To study the different methods of segmentation
2. Acquired blood images of normal and diseased individual.
58


Meeting Summary

MR. MUSTAFAIN ALI
Absent from
meeting:
Date: 12TH DEC 2019
Time: 11 AM
Points of Meeting:
1. To study the techniques of digital image processing through MATlab.
2. Show different methods of segmentation.
59


Meeting Summary

MR. MUSTAFAIN ALI
Absent from
meeting:
Date: 19TH DEC 2019
Time: 11 AM
Points of Meeting:
1. To study the methods of image enhancement in digital image processing.
2.
Next Meeting: 2ND JAN 2020 Signature: (Supervisor)
60


Meeting Summary

Absent from
meeting:
Date: 2ND JAN 2020
Time: 11 AM
Points of Meeting:
1. Detection of abnormality using artificial intelligence.
2. To study deep learning language for artificial intelligence.
Next Meeting: 9TH JAN 2020 Signature: (Supervisor)
61


Meeting Summary

Absent from
meeting:
Date: 9TH JAN 2020
Time: 11 AM
Points of Meeting:
1. To study neural networks.
2. To study YOLO(You Only Live Once) program for algorithm.
Next Meeting: 16TH JAN 2020 Signature: (Supervisor)
62


Meeting Summary

Absent from
meeting:
Date: 16TH JAN 2020
Time: 11 AM
Points of Meeting:
1. Applying different filters for image processing.
Next Meeting: 17TH FEB 2020 Signature: (Supervisor)

63
NED UNIVERSITY OF ENGINEERING & TECHNOLOGY,

KARACHI

Meeting Summary
Meeting attended SYEDA MUNAM ALI (BM-16010)

by: JAVARIA FATIMA (BM-16022)
Absent from
meeting:
Date: 8TH July,2020
Time: 12 PM
Location: ONLINE.
Points of Meeting:
3. Discussion about change in method/procedure of project.
Next Meeting: 14th July,2020 Signature: (Supervisor)
64

KARACHI

Meeting Summary

MR. MUSTAFAIN ALI
Absent from
meeting:
Date: 14th July,2020
Time: 12:15 PM
Location: ONLINE.
Agenda: Topic discussion Meeting No: 12
Points of Meeting:
2. Studying of Articles and research papers related to procedure.
Next Meeting: 21st July,2020 Signature: (Supervisor)
65

KARACHI

Meeting Summary

Absent from
meeting:
Date: 21st July,2020
Time: 12 PM
Location: ONLINE.
Agenda: Literature Review Meeting No: 13
Points of Meeting:
3. Studying of Articles and research papers related to procedure.
Next Meeting: 28TH July,2020 Signature: (Supervisor)
66

KARACHI

Meeting Summary

Absent from
meeting:
Date: 28TH July,2020
Time: 1 PM
Location: ONLINE.
Points of Meeting:
3. Discussion of selected method and work division among group members.
Next Meeting: 11TH August,2020 Signature: (Supervisor)
67

KARACHI

Meeting Summary

Absent from
meeting:
Date: 11TH August,2020
Time: 12 PM
Location: ONLINE.
Points of Meeting:
3. Finding data set of images containing Leukemia or no Leukaemia.
4.
68

KARACHI

Meeting Summary

MR. MUSTAFAIN ALI
Absent from
meeting:
Time: 12:30 PM
Location: ONLINE.
Points of Meeting:
3. Training of images under label of Leukemia or Leukaemia.
69

KARACHI

Meeting Summary

MR. MUSTAFAIN ALI
Absent from
meeting:
Time: 12 PM
Location: ONLINE.
Points of Meeting:
3. Training of images under label of Leukemia or Leukaemia.
Next Meeting: 2ND September, 2020 Signature: (Supervisor)
70

KARACHI

Meeting Summary

Absent from
meeting:
Date: 2ND September,2020
Time: 12 PM
Location: ONLINE.
Points of Meeting:
3. Showing the results to ma’am of labelled images and the point of accuracy
algorithm is giving.
Next Meeting: 9TH September,2020 Signature: (Supervisor)
71

KARACHI

Meeting Summary

Absent from
meeting:
Date: 9TH September,2020
Time: 1 PM
Location: ONLINE.
Points of Meeting:
3. Discussion for remaining things and some upgradation that project need in terms of
training of more images for more accurate results.
Next Meeting: 16TH September,2020 Signature: (Supervisor)
72

KARACHI

Meeting Summary

Absent from
meeting:
Date: 16TH September,2020
Time: 12 PM
Location: ONLINE.
Points of Meeting:
2. Discussion for report.
Next Meeting: Signature: (Supervisor)

Leukemia Detection Through WBC's Microscopic Blood Images Using Deep Learning Algorithm

Uploaded by

Leukemia Detection Through WBC's Microscopic Blood Images Using Deep Learning Algorithm

Uploaded by

Leukemia Detection Through WBC’s

Microscopic Blood Images Using Deep

Javaria Fatima BM-16022

Nisma Asif BM-16029

Ramsha Hafeez BM-16032

Bachelor of Engineering in Biomedical Engineering

Names Roll Numbers Signature

Details of FYDP members

Supervisor Co-Supervisor Advisor (if any)

Endorsement from CBM, NED University of Engineering and Technology

APPROVAL FOR SUBMISSION

Name: Madam Rehana Kouser

Department of Biomedical Engineering

NED University of Engineering and Technology.

Approved by Head of Department

Professor Dr. Ali Raza Jafri

Department of Biomedical Engineering

NED University of Engineering and Technology.

LEUKEMIA DETECTION THROUGH WBC’S

APPROVAL FOR SUBMISSION 4

1.1 Problem statement 16

2.1 White blood cells-leukocytes 17

2.1.1 WBCs Morphology 18

2.1.1.1. Normal WBCs 18

2.1.1.2. Abnormal WBCs 18

2.2.1 Types of leukemia 20

2.2.2 Leukemia recorded cases 22

2.3 Traditional leukemia testing 23

2.4 Artificial intelligence 24

2.5 YOLO – You Only Look Once 26

2.5.1 How does it work? 27

2.5.2 Implementation in detection 28

3. MATERIAL AND METHODS 31

APPENDIX I: Plagiarism Report 52

APPENDIX II: Log of meetings with supervisor 53

TABLE TITLE PAGE

FIGURE TITLE PAGE

Figure 1.1. Recorded leukemia cases globally by WHO 14

Figure 2.1.2. Leukocytes. (Micrographs provided by the Regents of University of Michigan

Figure 2.1.6. Hyper segmented neutrophil by imagebank | hematology| “Hypersegmented

Figure 2.2.2. Identified patients with different types of leukemia 22

Figure 2.4.3. Increasing use of AI in Pakistan by Z. Hoodbhoy, B. Hasan, and K. Siddiqui,

Figure 2.5.1. Subset of AI by S. Singh, “Cousins of Artificial Intelligence | by Seema Singh |

Figure 2.5.3. The working basics of our leukemia detector 28

Figure 3.1. Annotation for leukemia 34

Figure 3.2. Another annotation for leukemia 34

Figure 3.3. Annotation for no leukemia 34

Figure 3.4. Another annotation for no leukemia 35

Figure 3.5. Label files 36

Figure 3.6. Obj.name file 37

Figure 3.7. Obj.data file 37

Figure 3.8. Training file outlook 38

Figure 3.9. Methods used to train the model 40

Figure 4.1. First leukemia detection 41

Figure 4.2. Another detected leukemia 42

Figure 4.3. No leukemia detected 43

Figure 4.4. Another no leukemia detection 43

Figure 4.5. Error rate 44

LIST OF SYMBOLS / ABBREVIATIONS

WBC White Blood Cell

AML Acute Myelogenous Leukemia

ALL Acute Lymphocytic Leukemia

CML Chronic Myelogenous Leukemia

CLL Chronic Lymphocytic Leukemia

YOLO you only look once

k-NN K-Nearest Neighbor

SVM Support Vector Machines

APPENDIX TITLE PAGE

APPENDIX I: Plagiarism Report 53

APPENDIX II: Log of meetings with supervisor 54

Figure 1.1. Recorded leukemia cases globally by World Health Organization

2.1. White blood cells- leukocyte

2.1.1. WBC Morphology

2.1.1.1. Normal WBCs

2.1.1.2. Abnormal WBC

Some of the anomalies observed in WBCs are as follows:

resized_image = cv2.resize(image,(3width, 3height), interpolation = cv2.INTER_