In the doughnut-shaped buildings, particle accelerators take very detailed X-ray images. However, these images are not enough to learn how to drive on hydrogen, for example. PhD student Allard Hendriksen (CWI) has developed an algorithm that enhances images without having to learn data from previous measurements, because such data does not exist. He defends his doctoral thesis on March 3 at the University of Leiden.
Most impressive to Allard Hendriksen were the synchrotrons he visited in Switzerland and France. “I knew about the CT scans with the hospital X-rays, but they’re huge donuts in the landscape; that near Grenoble has a diameter of eight hundred meters. The electrons spin there at about the speed of light. fly out of one of forty corners, they shoot x-rays out of the ring.”
To run on hydrogen…
While a scanner in the hospital sometimes requires you to stand still for a few minutes, a synchrotron performs ten complete 3D reconstructions per second. That’s a billion times stronger than the hospital scanner. This is not enough to properly track a working fuel cell, for example. “Toyota wants to make hydrogen driving possible. For that, you need to be able to see exactly what’s going on in such a fuel cell while it’s running. We want up to three thousand frames per second.”
…To visualize each brain cell of a mouse
Sometimes the current power is sufficient, but the object under examination cannot withstand the radiation. “A piece of brain tissue swells up like a pudding from so much radiation. For example, researchers are working on a complete image of a mouse brain, with all the billions of nerve cells and all the connections between them. years, because now it can only be done by placing microscopically thin slices under the microscope, one by one.A gigantic project that creates a lot of data.
Algorithms usually learn from existing datasets
Hendriksen loves these large datasets. He has developed a self-learning or deep learning algorithm that can display images with the required detail with less radiation or with more speed. Normally, algorithms learn from existing datasets. Give them lots of brain scans and they’ll understand what brains look like at the level needed to distinguish, say, a tumor. This data on hydrogen fuel cells and brains at the cellular level is exactly what we don’t have.
This algorithm learns how not to
The solution found by Hendriksen seems a bit strange. “I let the algorithm learn how not to. I split the data set in half. With half a set, the algorithm has to try to find the other set. From one set to another and vice versa; we can always check how well it fits.”
It really works, and mathematically it’s correct
Once the algorithm has sufficiently filled in the gaps, you can trust it to correctly construct a twice as sharp, or even sharper, image based on the full scanner data set. By thinking mathematically about what his algorithm should do, Hendriksen was also able to reason theoretically as he could. His findings are now being applied to brain research at the cellular level.
Breakthrough: 299,924 instead of 300 photos per second
All sorts of data scientists gave Hendriksen datasets they had already tried everything with. “The best time was when I was working on a fuel cell dataset. It just wasn’t working. Until I started digging into the data. I was told that the speed of rotation was three hundred frames per second It turned out to be 299924. That was the breakthrough, now I figured out that the different slices of the 3D image just didn’t fit together and kept rotating slightly After correction, the image suddenly appeared.
Allard Hendriksen carried out his doctoral research in the Computational Imaging research group at CWI (the national research institute for mathematics and computer science in the Netherlands) and at LIACS (Leiden Institute of Advanced Computer Science). The thesis defense takes place on March 3, 2022. The thesis supervisor is Joost Batenburg. Check the University of Leuden events calendar for practical defense information.
Text on the thesis defense of Allard Hendriksen from the CWI: Rianne Lindhout
Photo: ESRF, Grenoble