GUJARAT TECHNOLOGICAL UNIVERSITY
A project on:
TUMOR DETECTION FROM BRAIN MRI
Project is under UDP
This is to certify that the project report entitled “TUMOR DETECTION FROM BRAIN MRI presented by KRISHNA PATEL, NISHIT SEHGAL, KAVISH GOENKA,HIMANSHU PATEL and SHRISH PATHAK bearing the enrollment numbers : 150050107043, 150050107054, 150050107505, 150050107506, and 170057107002.
PROF.ASHISH PRAJAPATI DR.AVANI VASANT
(Faculty Guide) (HOD)
Babaria Institute of Technology
Department of Computer Engineering
At: Varnama, Ta: Vadodara, Dist: Vadodara, Pin: 39124
We have taken great efforts in this project. However, it would not have been possible without the kind support and help of every individual and organization. We would like to extend our sincere thanks to all of them.
We are highly indebted to Babaria Institute of Technology for their guidance and constant supervision as well as for providing necessary information regarding the project & also for their support in completing the project, our project’s internal guide helped us a lot in this project and they are always there for our solution.
We would like to express our gratitude towards our parents ;members of Babaria Institute of Technology for their kind co-operation and encouragement which helped us in completion of this project. We would like to express our special gratitude and thanks to industry persons for giving us such attention and time.
Our thanks and appreciations also go to our colleague in developing the project and people who have willingly helped us out with their abilities.
The purpose of our project “Tumor Detection from Brain MRI Image” is to make people aware about the new technology and hence provide a technological buffer that helps people understand and process the current technology to benefit them in their day to day life.
This approach is made to make some improvements in addition with being updated to the new technology related to gesture controlling via mathematical algorithm. Here medical diagnosis through machine learning is considered to be important issues of artificial intelligence. Here in this project we present a machine learning approach to detect whether the MRI image of brain contains tumor or not. We are trying to make the design user friendly so that there will not be any problems while using. We hope that we would be successful in completing the project in a proper way.
1.1 Problem Summary7
1.2 Aims and Objectives8
1.3 Problem Specification9
1.4 Literature Review10
1.5 Plan of Work 11
1.6 Materials and Techniques Required12
Design: Analysis, Design Methodology and Implementation Strategy132.1 Observation Matrix (AEIOU Summary)132.1.1 Activities14
2.1.2 Environment 15
2.2 Ideation Canvas19
2.2.4 Props/Possible Solutions21
2.3 Product Development Canvas22
2.3.3 Product Experience25
2.3.4 Product Functions25
2.3.5 Product Features26
2.3.7 Customer Revalidation27
2.4 Empathy Mapping Canvas29
2.4.4 Story Boarding32
2.5.1 Use Case Diagram34
2.5.2 Activity Diagram36
2.5.3 E-R Diagram37
2.5.4 Class Diagram 38
2.5.5 Sequence Diagram 39
3.1 Modules in the System 40
3.2 Data Dictionary41
3.3 Flow Chart42
4.1 Advantages of the System43
4.2 Unique Features43
4.3 Conclusion and Scope of further Work44
Under the guidance of professor Ashish Prajapati, we the students of the Babaria Institute of Technology pursuing Computer Engineering are presenting our project “Tumor Detection From Brain MRI Image” which can be used as standard analytical tool for Brain Tumor Specialists around the world.
1.1 Problem Summary
If we talk about the number deaths that occur because of malignant brain tumor and other tumor then it will be in millions, here because of late diagnosis and false diagnosis the survival rate decreases, thus making it difficult to survive from these brain tumor diseases. The average survival rate for malignant brain is also low that is between 30-40%.
In this project we classified whether tumor is malignant or benign, size of tumor, category of tumor etc, which will help the specialists to take more informed decisions.
1.2 Aim and Objective of the project
Our Aim is to create an application which will take MRI images as input, segment the tumor in the image and transfer the segmented images to specialist’s id, these are analyzed by the specialists having all the case details via Web and Mobile application, this creates a directory of patients for the specialists, so that they can check it at any time via cloud.
Current manual tumor segmentation takes lot of time to segment out tumors in each individual slice in 3D image as a result of it effort increases.
Here the classification of tumor will be accurate if tumor segmentation is automated using deep learning techniques it will increase the efficiency and accuracy significantly, it will also reduces time and effort.
The statistical analysis of MRI images will also help the specialists to find details like tumor size etc, it will result into the accurate and precise treatment plans for patients.
1.3 Problem Specification
The objective of our current project is to remove/reduce the problems of the currently used method of segmentation i.e. manual segmentation performed by specialists.
The main problems of the current method are:
Prone to Errors
Not Very Accurate
Depends on Experience of the Specialist
Thus, to resolve the above mentioned problems we plan to propose a method for automatic segmentation of tumor from brain MRI using deep learning.
1.4 Literature Review
Review paper we referred to indicates four phases for effective detection of brain tumor from MRI images, it consists of four phases:
In these phase MRI input image is processed in such a way that it becomes suitable for the further processing.
In these phase image which contains unwanted noise is removed which gets added due to MRI scan.
These phase include to improve the appearance i.e. contrast and visibility.
In image enhancement Gaussian filter is applied to improve quality of image and removal of noise.
Segmentation In Segmentation EDPSO method is used for detection and segmentation of tumor from MRI images.
Neuro-fuzzy is sophisticated framework for multi-object classification. Fizzy rules are selected to classify an abnormal image to corresponding tumor type.
This technique is fast in execution, efficient in classification and easy in implementation.
1.5 Plan of work
To make the development process of our project fast, efficient and smooth, we have decided to divide the project in to a set of task which are as follow:
Surveying the cancer specialist hospitals to know in detail about the current problems in the manual method of segmentation.
Gathering the required training data-set for training our model from various hospitals and online open sources.
Finding a method to convert the 3D MRI images into 2D image slices and a way to represent them so that the specialists can view each image slice by slice.
Searching and testing suitable automated deep learning models such as BiSeNet, U-Net, SegNet, etc. to find a model which suits our requirements and gives best result with our current training data-set.
After finding the suitable model, we will develop the model and keep the model for training using the training data-set acquired from various sources.
After the training is over, we test the model using test data-set to check whether the model works as intended and give high accuracy.
Finally, we will consult various specialists regarding our project and take their suggestion to further improve the project and make it better for them.
1.6 Materials and Techniques Required
For the purpose of completing our project we require the following Materials and Techniques.
1. Training Data-Set
2. Testing Data-Set
3. Database Server
5. Cloud Based GPU Server for Training the Model
6. Validation Data-Set
1. Python Programming
2. Android Programming
3. Mongo DB
4. HTML & CSS
5. Node JS
6. Java Script
2. Design: Analysis, Design Methodology and Implementation Strategy
2.1 Observation Matrix (AEIOU Summary)
The AEIOU Summary canvas gives a brief description about the project such as what it will do (activities), where it will be used (environment), for whom is it useful (interactions), what materials are required to operate the project (objects), and who will use it (users).
Fig. 2.1 AEIOU Summary
The activities part of the canvas roughly describes what the project will do. Some of the activities performed by our project are:
1. Brain Tumor Segmentation
2. Image Slicing
3. 3D to 2D Image Conversion
Fig. 2.1.1 Activities
The environment part of the canvas roughly describes where the project will be used. Some of the places where our project will be used are:
1. Imaging Centers
Fig. 2.1.2 Environment
The interaction part of the canvas roughly describes for whom the project will be useful and who all will be interacting with our project. Some of them are:
Fig. 2.1.3 Interactions
The objects part of the canvas roughly describes what the project requires to work. Some of the objects required for the project are:
1. MRI Images
2. MRI Scanner
Fig. 2.1.4 Objects
The user’s part of the canvas roughly describes who will be using our project and who all will be interacting with our project. Some of them are:
Fig. 2.1.5 Users
2.2 Ideation Canvas
The ideation canvas roughly describes how the idea for the project came to be, i.e. for whom the project is, what the project will do, due to which situation did we get the idea for the project, and our possible solution to the problem.
Fig. 2.2 Ideation Canvas
It states for whom the project will be useful for, some of the people who will benefit from our project are:
Fig. 2.2.1 People
It states the activities performed by the project such as:
1. Tumor Segmentation
2. Image Slicing
Fig. 2.2.2 Activity
2.2.3 Situation, Context, and Location
This part of the canvas describes the situations and their context due to which we got our idea for the project
Fig. 2.2.3 Situation/Context/Location
2.2.4 Possible Solutions
This part of the canvas describes our solutions to the above discussed problems.
Fig. 2.2.4 Props/Possible Solutions
2.3 Product Development Canvas
The product development canvas give more specific information related to the technical part of the project such as:
Fig. 2.3 Product Development Canvas
This section states the purpose of the project; why we are making the project.
The purpose of our project is to provide a method for automated brain tumor segmentation using deep learning which as fast, efficient and accurately.
Fig. 2.3.1 Purpose
This section gives the information about the people who are going to use the project and also the information on the people who are going to benefit from the project such as:
Fig. 2.3.2 People
2.3.3. Product Experience
This section gives an idea about how the people using the product will feel while using this product. We aim to make this product
2. Easy to Use
3. Fast and Accurate
Fig. 2.3.3 Product Experience
2.3.4 Product Functions
This section describes in brief what the functions of the product are:
1. Brain Tumor Detection
2. Brain Tumor Segmentation
3. Maintain Patient Database and Report
Fig. 2.3.4 Product Functions
2.3.5 Product Features
This section describes in brief the features of our products which will be:
1. Cross Platform Usable
3. Friendly U.I.
Fig. 2.3.5 Product Features
This section describes in brief the components/techniques used in the product. The techniques which we will be using for our product are:
1. Python Programming
2. Node JS
3. Android Programming
Fig 2.3.6 Components
2.3.7 Customer Revaluation
This section gives a brief of the customers` feedback after using a similar product/ hearing about our ideas. Some of the feedbacks are:
1. Increasing Accuracy
2. Increasing Precision
3. Making U.I. More User Friendly
Fig. 2.3.7 Customer Revalidation
2.3.8 Reject, Redesign, Retain
This section gives a brief of the features and functions we have Rejected / Redesigned / Retained after hearing the customer revaluation. Some of the changes made after the revaluation are:
1. Improved Post Processing
2. Increased Data-Set
3. Changed Model for Better Accuracy
Fig. 2.3.8 Reject, Redesign, Retain
2.4 Empathy Mapping Canvas
The empathy mapping canvas discusses the users, the stakeholders related to them, the activities performed by the project and also few stories which we heard related to our project from the people who are going to use the product.
Fig. 2.4 Empathy Mapping Canvas
This section mentions the users who are going to use and benefit from the product such as:
Fig. 2.4.1 User
Stakeholders are the people who are indirectly affected due to the project such as:
1. Patients` Families
Fig. 2.4.2 Stakeholders
These are the activities performed by the project such as:
1. Image Slicing
2. Tumor Detection
3. Tumor Classification
Fig. 2.4.3 Activities
2.4.4 Story Boarding
In this section, we talked to many people related to the project such as patients and their families, doctors, radiologists, neurologists and they told us some stories related to the project; some of which were happy stories while some sad stories
184.108.40.206 Happy Stories
1. Kamlesh is a milkman, he was diagnosed with brain tumor, but because of automated brain tumor detection tool in hospital, his life was saved because of fast and accurate treatment
2. There is one specialist in one of renowned hospital, as hospital renowned there are many number of cases, so for accurate and precise treatment he has to examine all the MRI image carefully and these take a lot of time, so with the help of automated brain detection tool not only the speed of examination increase but also accuracy and precision of report also increases.
Fig. 220.127.116.11 Happy Stories
18.104.22.168 Sad Stories
1. Kailash was a farmer, he was diagnosed with brain tumor but due to scarcity of brain tumor detection tools and specialists in the village hospital, he was unaware of the disease and later when he went to city hospital for diagnosis it was already too late to cure the tumor and he ultimately died.
2. In one hospital, because of large number of MRI images to be segmented, quick segmentation with high accuracy became difficult for the radiologists, as a result the life expectancy decreased because of the delay in treatment.
Fig. 22.214.171.124 Sad Stories
2.5.1 Use Case Diagram
A use case diagram is a representation of how a user will interact with the system. It shows the relationship between the user and the different uses of the application in which the user is involved.
Fig 126.96.36.199 General Use Case Diagram
-3175-227330Fig 188.8.131.52 Technical Use Case Diagram
2.5.2 Activity Diagram
-6985889635The activity Diagram is used describe how the system will be used and when it will be used. Activity diagram is a type of flowchart to represent how the system transitions from each activity to the next activity. The activity can be interpreted as a function of the system.
Fig 2.5.2 Activity Diagram
2.5.3 E-R Diagram
An entity relationship diagram (ERD) shows the relationships of entity sets stored in a database. An entity in this context is a component of data. In other words, ER diagrams illustrate the logical structure of databases.
Fig 2.5.3 E-R Diagram
2.5.4 Class Diagram
A class diagram is a representation of the relationships and source code dependencies among classes of the system. In this context, a class defines the methods and variables in an object, which is a specific entity in a program or the unit of code representing that entity.
Fig 2.5.4 Class Diagram
2.5.5 Sequence Diagram
Sequence diagrams, also called event diagrams or event scenarios are a series of parallel vertical lines (lifelines), different operations or users that operate simultaneously, and, the horizontal arrows are the information exchanged between them, in the order in which they occur.
Fig 2.5.5 Sequence Diagram
This Section details about the implementation related to the project.
3.1 Modules In The System
The project which we are aiming to complete will have 3 Modules which are as follows:
Front End Module
Back End Module
3.1.1 Front End Module
This module will be comprised of the front end aspects related to the project such as web page, android application, and web application.
The front end module also contains the method for converting 3D MRI images into 2D image slices which can be viewed by radiologists slice by slice to better study the MRI.
3.1.2 Back End Module
This module comprises of the back end data base which consists of the patients MRI records and the result of their segmentation along with the treatment process information.
3.1.3 Segmentation Module
This module will take a MRI image as an input and after applying the necessary segmentation on the MRI, it will give an output with the segment Brain MRI in the form of 2D image slices
3.2 Data Dictionary
Data Dictionary is a document describing a database schema for users. In most basic case this documentation includes descriptions of tables and their columns
Database Column Nu. Column Name Data Type Null
Patient Name Varchar Yes
2 Segmentation Result Varchar No
3 Segmentation Images Image No
4 Procedure Performed Varchar No
5 Neurologist Name Varchar Yes
6 Hospital Name Varchar Yes
Table 3.2 Data Dictionary
14744705410203.3 Flow Chart
Fig. 3.1 Flowchart
4.1Advantages of the System
As we know by now that manual segmentation of an MRI image is a tedious task and takes a lot of time thus making the method inefficient, and along with the segmented parts of the MRI, doctors can plan out efficient recovery paths by considering relevant analytical data but manual segmentation does not provide any relevant analytical data for the doctors to consider. When taking real life situations into account a single radiologist might take multiple MRI scans throughout a single day, thus this additional time required for manual segmentation adds up. This additional time can instead be used to take even more MRI scans.
One of the underrated advantages of the system is that it transfers all this medical data to the cloud with state of the art security, thus making the system secure and more accessible when compared to physical repositories of data.
Thus the automated segmentation method using deep learning is not just efficient but also offers analytics that can possibly change the space of brain tumor research for the upcoming future.
4.2 Unique Features
Our product provides analytical data like Volume of varieties of tumors
Rate of growth with multiple MRIs of the same patient
Faster and more efficient than manual segmentation
Combines medical expertise with the advantages of data analytics
Minimizes manual labor and large repositories by saving all the data in the cloud
Easily accessible on demand
4.3 Scope of further work
The methodology used for this project can further be improved and can be applied in a variety of fields in medical science like surgery and bone damage repair and in the future with a large enough dataset, this model might even cluster multiple cases and suggest possible treatments that have worked in the past for various other cases.