Code covered by the BSD License  

Highlights from
Brain Message v 1.0 to write secret messages into fMRI

1.0

1.0 | 1 rating Rate this file 12 Downloads (last 30 days) File Size: 14 MB File ID: #41062
image thumbnail

Brain Message v 1.0 to write secret messages into fMRI

by

 

Write secret messages in fMRI data to be uncovered with independent component analysis

| Watch this File

File Information
Description

Independent Component Analysis is a data-driven method to uncover independent signals in functional MRI data. These signals encompass a mixture of true functional networks, physiological and mechanical noise, and other artifact. The blob-like nature of many of these components sometimes reveals interesting shapes. So what if one day you performed the analysis and found a secret message?

Inspired by V1 visual cortex that sometimes looks like a heart, and the occasional component map that looks like the Koolaid man, this script makes it possible to write hidden words in functional MRI data to be uncovered with independent component analysis (ICA).

It takes as input a specified message, and an output file name, and outputs a nifti file for decomposition with ICA. Usage is as follows:

dummy = BrainMessage('message','output.nii');

You will want to create the image using the command above, and then run ICA. I normally use FSL’s melodic, however another good option is the GIFT toolbox, for those users who prefer Matlab and SPM. The script uses SPM to read and write images, and you can of course change these function calls to your preference. When running ICA, you can estimate the number of components or use your method of choice, however 46 should work best.
 
How does it work?

I start with a set of real component timecourses and spatial map distributions, and modify the spatial maps to take the shape of letter templates. The algorithm solves for the matrix W in the equation S = W X, where S is the observed signal, W is some matrix of weights, and X is the unmixed signal. Applied to fMRI, our matrix of weights consists of columns of timecourses, and the unmixed signal has rows of component spatial maps. I therefore went about this backwards, and multiplied rows of timecourses with edited spatial maps to come up with the observed data, S. In other words, for each letter, we multiply a column of Nx1 timepoints by a squished spatial map (a row of 1xV voxels) to result in a matrix of size NxV. I then decided to add in some real functional network spatial maps to make the data more realistic.

Where do the templates come from?

These three sets of templates come zipped up with the script, and there are 46 of them, meaning that you are limited to creating messages 46 characters long. Why did I choose to do this? I wanted this data to be as “real” as possible as opposed to the other option of generating random timecourses and distributions. You could easily edit the script to make everything faux and remove this limit. Secondly, I wanted uncovering the letters to not be consistently easy. By using real data I would be sampling from components with varying strength, and so a resulting spatial map can come out weaker by chance, making seeing it more challenging.

What can I change?

You can easily modify the script to generate fake timecourses, and not limit to messages of 46 characters. You can also create new templates (smiley faces, custom logos), or improve the code however you see fit. If you do something interesting, please share, and most importantly, have fun!

The scripts and a few completed example files are also available on github:

https://github.com/vsoch/BrainMessage

MATLAB release MATLAB 8.0 (R2012b)
Other requirements SPM
Tags for This File   Please login to tag files.
Please login to add a comment or rating.
Comments and Ratings (1)
09 May 2013 jiangfen wu

this programe cant run,it will hault at "Adding custom signal to data...",why?how to solve it?

Contact us