MRI is an imaging method of the brain, which is based on the magnetic resonance phenomena.
Abstract
MRI is an imaging method of the brain, which is based on the magnetic resonance phenomena.
Functional Magnetic Resonance Imaging or better known as fMRI is a very important and known method, which assists scientists to understand how the human brain functions.
An fMRI image (in the BOLD technique) shows the blood oxygenation level in different areas of the brain, where high oxygenation level implies an enhanced brain activity in the relevant area.
The goal of fMRI experiments is to find the areas (sources) in the brain, which are responsible for a certain activity.
Under the appropriate assumptions, finding the sources from fMRI data sequence can be treated as a Blind Source Separation (BSS) a well known problem in the field of signal processing.
This project deals with different ways for solving the BSS problem using sparse representations.
Further more, the project tests the implementation of the BSS solution on real fMRI data sets.
The problem (or the background)
We receive a real fMRI data set. The set consists of many images taken along a certain time course.
The images consist of several sources, which are activated in the brain during the time of the experiment.
Our goal is to separate all the sources and distinguish them.
The algorithm
First we preprocess the images – this includes removing the background, using gradient or laplacian on the images, dividing the images to several blocks or using Wavelet Packet.
Then, we try using two different algorithms in order to separate the sources:
FCM – a geometric algorithm, which can classify a set of coordinates to several groups
ICA – an algorithm which can separate the principle components of a given image.
The solution
We discovered that the ICA algorithm together with a gradient and Wavelet Packet preprocess gets the best results.
Conclusions
1) Visual Cortex activity
2) “Planted” sources that were separated
Acknowledgment
We are grateful to our project supervisors Michael & Alex Bronstein for their help and guidance throughout this work.
Also we would like to thank Dr. Michael Zibulevsky and Mr Yohanan Erez for their help.
We are also grateful to the Ollendorf Minerva Center Fund for supporting this project.