Brain Cancer Detection using MRI Images

Brain tumor is one of the deadly diseases with the least chance of survival. Doctors diagnose brain tumor through manual processes in which the tumor is identified, classified into tumor type, and then segmentation of the brain tumor is performed. The manual process is difficult and time-consuming,

Project Title

Brain Cancer Detection using MRI Images

Project Area of Specialization

Computer Science

Project Summary

Brain tumor is one of the deadly diseases with the least chance of survival. Doctors diagnose brain tumor through manual processes in which the tumor is identified, classified into tumor type, and then segmentation of the brain tumor is performed. The manual process is difficult and time-consuming, while early tumor detection takes time. In order to provide adequate treatment to patients in a timely manner, brain tumor segmentation plays an important role in medical imaging, so automating these manual processes will save the doctors’ time and chances for patients’ survival will also increase. We have thoroughly reviewed literature from 2019 to 2020 for these types of brain cancers: Glioma, Meningioma, Pituitary from different databases and search engine including IEEEXplore and Google Scholar, and found 16 papers which use deep learning techniques to perform classification and segmentation. To the best of our knowledge, there are no mobile applications and web portals easily available that could help the doctors in classifying the type of tumor and performing segmentation automatically using Magnetic resonance imaging (MRI). In order to fill this gap of knowledge, deep learning model will be used to automate the tumor segmentation and classification processes, and then the web portal and mobile application will be developed for the analysis of brain tumor timely and efficiently.

Project Objectives

The gap related to brain cancer segmentation and detection using MRI images has been found after thorough literature review as mentioned above in problem identification section. There has been done a lot of research and practical work on brain cancer detection. Different models have been proposed and transfer learning has been applied on the pre-trained model to bring up solution to brain cancer detection as it is the need of time. Better solutions with highest accuracy are most needed for early detection of brain cancer to save lives of people. To the best of our knowledge, web application and mobile application or end-to-end systems based on MRI image for the detection of cancer are not available. Therefore, to fill this gap we will work on these two domains to produce workable product that could target large population and will be accessed anywhere any time. The application will make use of convolutional neural network for the brain cancer detection and machine learning for cross-validation. The model will be trained on the Kaggle brain cancer dataset based on MRI images which will first classify images to be cancerous or non-cancerous and then output of the CNN will either be classified into one of the three classes as glioma, meningioma, pituitary or proceed further for cross validation using machine learning techniques. In the context of above conversation, our project could be summarized as:

1. To study and evaluate the pre-trained GoogleNet CNN model.
2. To study and examine the Support vector machine (SVM) classifier.
3. To apply cross-validation on results of CNN model that initially predicts the MRI images as healthy.
4. To apply segmentation on the brain MRI images to detect tumor region in brain.
5. To apply data augmentation (flipping, rotating etc.) to increase the data size and make the model feel that these are different images, which will help increase the accuracy.

Project Implementation Method

Data Augmentation is a method to increase the size of training data by creating different versions of the same image in order to increase the performance of convolutional neural network(CNN) for classification into three type of cancers (Glioma, Meningioma and Pituitary) is there is no any tumor in the MRI Image to decrease the chances of false result cross validation will be performed using machine learning algorithm support vector machine (SVM) after classification result will be cross validated

Benefits of the Project

Brain cancer being part of medical imaging is getting importance day to day, in recent ten years, medical field has appeared as one of the emerging technologies. From our thorough literature review, we have identified an unmet need that there is no availability of web portal and mobile application that operates on the MRI images and classify the brain cancer into one of the three classes such that Glioma, Meningioma, and Pituitary cancers which could help vast population. There are different web applications and mobile applications that are providing information regarding brain cancer, but they do not classify brain cancer. Subsequently, there are also desktop applications in this domain that classify brain MRI to be either cancerous or non-cancerous, and classify the MRI images into Glioma, Meningioma, and Pituitary brain cancer. But the problem is that the desktop applications are not available to all the people, but just to the targeted physicians and neurosurgeons.

Target customers are the population for whom a solution to specific problem is being designed. Target customers include doctors, clinicians, patients, caretakers, families, and researchers, all being primary stakeholders.

Technical Details of Final Deliverable

Web Application

The Web Application can detect and will be able to perform classification and segmentation using flask framework for model training and php will be used to develop frontend of the application and will be deployed on localhost for demonstration[Figure 2].

Figure 3 Web Application Architecture

Mobile Application

The Android application can detect and perform classification and segmentation of three type of brain cancers (Glioma, Meningioma and Pituitary) using MRI scanned images all the models are trained on publicly available datasets from Kaggle. I have used python to train the models and TensorFlow lite to convert into mobile based system to run natively on mobile devices and to built a mobile application using android studio.

Tools and Technologies

Tools and technology that we will be using are listed as:

• Python
• Tensor flow (Tensor flow lite for mobile application)
• Flask Framework for Web Application
• Android Studio Java for Mobile application

Final Deliverable of the Project

Software System

Core Industry

IT

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com