Design of a Computer Assisted Diagnosis System for Brain Tumor Detection with FPGA Implementation

Brain-Tumour diagnosis requires the patient to undergo scanning (usually MRI scanning), and the doctors diagnose the location and type of tumour, by looking at the scans. There is almost 30% chance of misdiagnosis due to human error, the small size of the tumour, etc. To prevent this misdiagnosis, t

Project Title

Design of a Computer Assisted Diagnosis System for Brain Tumor Detection with FPGA Implementation

Project Area of Specialization

Biomedical Engineering

Project Summary

Brain-Tumour diagnosis requires the patient to undergo scanning (usually MRI scanning), and the doctors diagnose the location and type of tumour, by looking at the scans. There is almost 30% chance of misdiagnosis due to human error, the small size of the tumour, etc. To prevent this misdiagnosis, this project aims to develop a system that will assist the doctors in the proper diagnosis of the brain tumour in terms of its presence, localization, and type.

This implementation will use Deep Learning technology. A Deep Learning model consists of multiple layers that represent data with multiple levels of abstraction.

The convolutional neural network is a powerful method for image recognition and prediction. It will be used for brain tumour segmentation, classification and localization of the tumour. 

The benefits of CNN is that it is designed to determine features adaptive through backpropagation by applying numerous building blocks such as convolutional layers, and fully connected layers. The 3D Convolutional Neural Network consists of multi-channel metric maps that are used to extract the high-grade predictive features from the individual patch of these maps and trains the network layers for prediction.

This diagnostic algorithm will be implemented using MATLAB and FPGA for using it as a stand-alone system, which would be easier to use for the doctors. 

High-grade gliomas brain tumour is very aggressive and leads to the death of a patient in 1 to 2 years that is why the accurate and timely diagnosis is necessary which increase the survival time of a patient.

The customized dataset, which is made from two datasets publicly available contains, T1 enhanced brain MRI images (512*512 axial images). Datasets are collected from:

1. Kaggle (for detection of tumour).
2. Figshare (for classification of tumour).

Project Objectives

The following are the objectives to be achieved:

1. To detect the presence of the tumour.
2. To segment the region of interest from an object and segmenting the tumour from an MRI Brain Image.
3. Accurately classify a type of tumour from 2D Brain MRI Images.
4. To implement the designed algorithm on FGPA (Field Programmable Gate Array) for commercialization.

Project Implementation Method

Design Of A Computer-Assisted Diagnosis System For Brain Tumor Detection With FGPA Implementation.

For an accurate diagnosis of brain tumour, Convolutional Neural Network is used. Convolutional Neural Networks (CNNs) are a unique deep learning structure originally modelled on a human virtual cortex. The reason for using a convolutional neural network is that it provides optimal accuracy of segmentation.

Due to this, the processing time decreases and the tumour can be located in the MRI Scans. And if the tumour is present there than we can also detect the type of tumour. And we can also predict the life expectancy of that patient.

Steps to be followed:

• Two datasets are collected from Kaggle and Figshare.
• Normal MRI Scans are collected from Kaggle and the diseased MRI Scans are collected from Figshare.
• Two Dataset are combined to form customized dataset.
• This customized dataset is divided into the training set, validation set and testing set.
• These three sets are sent to CNN which is a Pretrained Network with Transfer Learning.
• CNN is pre-trained with AlexNet, GoogleNet and ResNet.
• After this, all process output is produced.
• If the output is validated (confidence level must be greater than 95%) than diagnosis occurs.
• If the output is not validated (confidence level is less than 95%) than it is sent back to CNN after hyperparameter tuning.
• The MRI scans are then sent to any specialist or diagnostician for further treatment.

Benefits of the Project

Some of the benefits that can be achieved after implementing this project are:

1. Patients with the tumour can get accurate treatment on time, due to proper diagnosis and it reduces decision making time.
2. Detection of the tumour will be more accurate as compared to just a single specialist's/diagnostician's diagnosis.
3. Design of a hardware system for the real-time diagnosis will be commercialized for use in radiotherapy, chemotherapy and operative treatment planning.

Technical Details of Final Deliverable

At the end of the project, we expect to have a system which will be given T1 MRI scans of the Brain and will give the indication whether there's a tumour or not, and if yes, then the location and type of tumour as the output.

Following are the final project's deliverables:

1. An algorithm for Tumor Detection and Diagnosis/Classification.
2. A ready-made hardware system as a real-time diagnosis system.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Health

Other Industries

Education , IT , Medical , Others

Core Technology

Artificial Intelligence(AI)

Other Technologies

NeuroTech, Big Data

Sustainable Development Goals

Good Health and Well-Being for People, Quality Education, Decent Work and Economic Growth, Industry, Innovation and Infrastructure, Reduced Inequality, Life on Land, Peace and Justice Strong Institutions

Required Resources

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