Medical Imaging and the Wondrous Robot Uprising

April 14, 2026

You’ve just gotten an MRI.

Most of the resulting scan is clear; however, the image quality is…suboptimal. Even with proper protocols to minimize issues, the machine, the software, and the patient’s demographics and implants can all lead to noise, artifacts and other irregularities.

Imaging in science fields is rife with such situations. From medicine to research, scientists are going to want a clear picture of their subject of interest. So, how do you solve this problem?

You use math! There are all kinds of special algorithms in use, such as image reconstruction algorithms and metal artifact reduction techniques for dealing with metal implant-induced artifacts. And with new breakthroughs in artificial intelligence, there are now new ways to boost image quality.

To learn, artificial intelligence can either use a model-free method (which relies on positive and negative reinforcement to incentivize the program towards a specific outcome) or a model-based method (which relies on using mathematical models for decision-making). Model-free methods are faster, but are less suited when more discrete traits such as age of onset of a disease are involved.

Model-based methods train by going through multiple iterations of mathematical models and using them to predict which outcome is best.

There are three strategies used for their training:

The plug-and-play method integrates pre-existing AI models and currently sees lots of use in medical imaging.
Deep unrolling methods “unroll” each iteration into different layers, and each subsequent pass through a layer improves the next layer. These methods have very high accuracy when running up to the number of iterations they have been trained on. Past that point, though, the output can get a bit…unstable.

The newest and most exciting approach is the deep equilibrium method. As iterations of a deep unrolling method continue, they tend to converge towards a fixed point. By calculating that fixed point and choosing that as the endpoint, you can optimize the process by saving on memory consumption. This method is less memory-intensive and scales well to larger tasks. Gilton et al. were the first to use standard fixed-point accelerators to speed up calculation of the fixed point, further optimizing the process.

These new advances in A.I. ensure that even an imperfect MRI can yield results both useful and accurate enough for a diagnosis…as long as the model is trained properly, of course.

Storing Data on...Glass?!

March 6, 2026

Glass is rather notoriously the poster boy for fragility. It shatters, it breaks, it cracks. Mr. Glass from the movie Unbreakable is very much NOT the titular unbreakable character. And glass jaw isn’t winning you any boxing matches.

Counterintuitively, Microsoft thinks it can be the ultimate long-term storage method. Enter Project Silica.

See, glass might not be shatterproof, but it has a ton of other traits. It’s chemically inert, and nonreactive to electromagnetic frequencies, which makes it weirdly durable. How durable?

Media stored on glass can last 10,000+ years, according to Microsoft. This could revolutionize long-term data storage.

Something you might initially think wouldn’t last a decade could store data for a dozen millennia. Science can be so weirdly cool sometimes!

A Worm That Can Be Cut In Half!

June 22, 2025

You may have heard that one rumor about cutting earthworms in half. That, if you split ‘em straight down the middle, they’ll grow into two new earthworms?

Yeah…that’s just a myth. Please don’t do this.

Earthworms can regenerate their tail end if it’s cut off, but they still won’t regenerate into two separate worms. Damage from the clitellum (that larger segment in the middle) and above will kill the worm.

That said, while earthworms can’t be split into two separate worms, that doesn’t mean we the rumor is 100% false! There’s a kernel of truth to it…we’re just looking at the wrong worm. We need to turn our attention towards these little guys: an order of flatworms known as Tricladida…or more commonly, planarians.

Many flatworms are parasitic, but planarians are among the kinds that aren’t! Some of ‘em are even quite cute, with little rudimentary eyes and an adorably squooshy body. But that isn’t the only cool thing about them. They have amazing regenerative abilities, like recover-from-the-tiniest-fraction-of-their-bodies level. And yest, if split right down the middle, they’ll regenerate into two new li’l guys.

One more thing: if a planarian is only partially split, it’s possible for you to get one with two heads, as each half regenerates its own head! Kinda makes me wonder how their digestive system works in that case. Do they share a single digestive system? Considering how planarian digestive systems work, that could have some weird ramifications. Normal planarians don’t have an anus, they just have a single “blind sac” that both takes food in and excretes waste. So…the worst case scenario for our two-headed friends is…

…Probably best not to think about it.

What's New In Science: Peanut Allergies

June 12, 2025

There has been a fascinating study on peanut allergies published in Science Translational Magazine this month, and it could have huge significance towards our understanding of how we develop allergies in the first place. But in order to understand all that good stuff, here’s what you’ll need to know about antibodies:

Antibodies are molecules produced by a type of white blood cells known as B cells.

All B cells are the same at first, and share the same basic DNA (“germline sequences”, if you’re feeling fancy). When B cells begin to mature, that B cell’s DNA rearranges itself. Because of that rearrangement of DNA, the antibodies that B cell (and its “kids”) produces will be completely unique to the antibodies other B cells produce. That antibody will randomly have an affinity to a very, very specific random particle, called an antigen. B cells aren’t big believers in being a “Jack of All Trades”.

When that antigen is a viral particle, everything is just dandy. The immune system is working as intended.

When that antigen is a particle found in peanuts…then you have a problem.

Which brings us to the study in question.

The researchers looked into different antibodies that all reacted to the same common peanut antigen. Despite having completely random DNA rearrangements, the B cells all came to the same conclusion: “PEANUTS MUST DIE!!!!”. And now their respective humans need to carry an Epi-Pen for the rest of their lives.

So, our researchers, Marini-Rapaport et al, asked the obvious question: what the HECK is going on with those DNA rearrangements to make peanut allergies so danged common?

First, they identified exactly what parts of the antigen the antibodies all interacted with. They found that the DNA that codes for those interactions was found in base DNA, pre-rearrangement. That makes it more likely that, when recombination occurs, the peanut-binding portion remains.

In other words, peanut allergy isn’t just a purely random thing: our deck is metaphorically stacked against peanuts for some unknown reason! For all we know, ancient peanuts could have had a heck of a kick, for human evolution to see it and go “yeah, I’d rather die than eat THAT again”.

White Blood Cells as Pokémon

June 12, 2025

Something I loved about white blood cells as a kid was how similar to Pokémon they can be. They come in different types, can “evolve”, and even have special moves and abilities they use! Just for fun, here are the five major “species” of white blood cells, reimagined as Pokémon: