Classification of fMRI data for diagnosis of neurodegenrative diseases.

Introduction

In their paper, Whitfield-Gabrieli et al. show how the differences in brain activity of schizophrenia patients and healthy individuals can be monitered usind fMRI. We intend to use such fMRI data for finding out the differences in brain activity of healthy individuals and patients of neurodegenerative diseases and design a classifier that could be used for diagnostic purposes. In this document, we first discuss Whitfield-Gabrieli's work and then present our project proposal.

The Experiment

Functional MRI images of the brains of healthy individuals (referred from now on as controls) and schizophrenia patients and their first degree nonpsychotic relatives were taken during times when the subject was:

The part of the brain that was activated during the resting state was called the default network.This region would correspond to that part of the brain involved in task-independent function.This region most consistently included the medial prefrontal cortex (MPFC) extending to ventral anterior cingulate cortex, the posterior cingulatecortex (PCC) extending to the precuneus, and lateral parietal cortex. The activation of this default network was suppressed when the subjects performed working memory tasks and the amount of suppression was positively correlated with the healthy subjects' performance in the task. This has been shown in the figure below:


Source: Whitfield-Gabrieli et al. 10.1073/pnas.0809141106

 

It was, therefore, hypothesized that the default network would be hyperactive in patients with schizophrenia to the extent that it would be based on genetic risk factors and that this would be so in case of relatives as well.

WM task description:

Results

The control group performed better in the difficult WM task, but the performance of the patients and relatives was not significantly different from the controls in the easy WM task. The results are shown in the table below:


Source: Whitfield-Gabrieli et al. 10.1073/pnas.0809141106
 

The task related suppression was significantly less in patients and relatives than in controls as shown in the figure below:


Source: Whitfield-Gabrieli et al. 10.1073/pnas.0809141106
 

Discussion

As has been shown in the experiment above, techniques such as fMRI can be used for identifying regions in the brain that play a critical role in neurodegenerative diseases. Data from experiments like these is being used for diagnostic purposes as well. Using machine learning techniques, one can design fairly accurate classifiers that would be used for diagnosis of such diseases.
In our project, we intend to design such a classifier. We are trying to acquire data fMRI from actual patients and controls for this purpose. However, if by chance we fail to get such data, we intend to use simulated data of brains with multiple sclerosis, which is freely available here.

References