First of all, I must say I'm shocked: once again, we face an example of human collective idiocy, a paralysis of social structures embedded in the technology.
Somewhere, in the internals of the MDI embedded machine, the MDI image recognition algorithm builds a model of a part of human body, such as a head. The MDI algorithm, mathematically, is one of the most sophisticated procedures we can find in any products of technology produced by industry, probably the most sophisticated in the what is produced by the industries of the whole humankind, ever. And on the output, the imaging data comes out with the embedded text data generated using patient data manually entered be the personnel. It is apparent that such data exists before embedding of this text data, it is not made accessible in a "legal" way, just by reading it. At least for some of people involved in a technological chain.
It does not look like a hack used by a single company. According to the article, this hack is done by the authors. Let's see.
Reading of the article leave impression of deep illiteracy of the authors and their desire to look "scientific". The article is devoted to de-noising, and the "tracking algorithm" has nothing to do with it. They call "tracking algorithm" something they use to remove that very embedded text artifact the call "film artifact". Every one who ever called about MRI and what is called "tomography" knows that there is no such thing as a "film", ever. OK, let's dismiss it as a "professional" jargon.
The real problem is that, according to the description, there is no any "tracking algorithm". Throwing out strange term "flag value", it looks like the authors simply replace white pixels with black pixels, pixel by pixel, in hope that the brain image itself is only composed with "gray" pixels (not of the value 255), so the text artifacts are "recognized" due to extreme contrast in the text on black background compared to less contrast in the "useful" image.
This "algorithm" is highly unreliable. The text would remain clearly readable (and detectable by simple contrasting algorithm) if background color is slightly different from the value of 0 (say, 1-2). Worst, the algorithm, when applied to whole image, can damage some pixels in the useful part of image.
Brain damage!
I might still be applicable to most or all cases though, just try to use it. I would prefer manual selection of the areas to be cleaned up, but in this case, no algorithm is needed at all. There are more sophisticated algorithms which can do this job more reliably, and using them is a whole science. You can find them:
http://en.wikipedia.org/wiki/AForge.NET[
^],
http://www.aforgenet.com/[
^],
http://en.wikipedia.org/wiki/OpenCV[
^],
http://opencv.willowgarage.com/wiki/[
^]. In complexity, all such algorithms are nothing compared to MRI algorithms.
Most image recognition projects are commercial.
—SA