rasmus1610

yes. At least they want to use it for that, I am not sure they already do.

(look under "multimodality" in the blog post: https://thinkingmachines.ai/news/introducing-inkling/)

some may argue for ghostty but yeah :)

but those are two different things. Of course something like Deep Resolve is great, as are modern model based reconstruction algorithms for CTs, but here we are talking about LLMs and their ability to interpret medical images, which has nothing to do with what you said.

As a radiologist I have found Claude and ChatGPT to be absolutely terrible at MRI and I would not trust it one bit. It has its merits if you need to research stuff that is more text based, but radiological images is just something that they cannot interpret good enough (yet)

we don't even need AI for that: https://3dqlab.stanford.edu/what-is-volume-rendering/

It will be terrible for brain imaging. The ultrasound waves can’t go through the skull and thus can’t image the brain. Additionally you would have to drown the patient since you need a medium other than air between the ultrasound emitting probe and the body which is water in their device.

CT is more than sufficient for imaging the brain in a case of trauma and MRI is not automatically better than CT in every case.

(I am a neuroradiologist)

it is more like that the brain learns to use other regions or neurons to do the tasks of the dead brain cells. The brain cells that are dead due to ischemia are dead and will usually be collected by microglia and after some time there are defects in the brain where the ischemia was.

This is such an exciting direction :)

Jeremy Howard from fast.ai/answer.ai also works on similar stuff with solveit (https://solve.it.com) and ipyai (https://github.com/AnswerDotAI/ipyai)

I think it will be very interesting to see what this enables

I mean what is the alternative here?

Mario maintains the project pro bono, which is only possible because he had an exit a couple of years ago. I believe it is a lot of work making sure that the quality of the codebase doesn't detoriate with the onslaught of slop that is hitting open source projects right now.

Additionally, I have massive respect for Armin and I don't believe Earandil is your typical VC startup that wants to grow no matter what.

Maybe Chris Lattner leaving and creating Mojo also didn’t help in that regard.

Swift for TensorFlow was a cool idea in that time …

Hey HN, I'm Marius, an AI researcher and radiologist.

I read a lot of ArXiv papers on my Kindle Scribe, but other tools for sending PDFs to Kindle completely scramble math equations and tables.

I built PDFling to bypass that. It just takes the ArXiv URL and sends the raw, unharmed PDF directly to your device via Whispernet.

It's a simple scratch-my-own-itch project, but I figured others here might find it useful.

I'd love to hear your feedback :)

So many people are so salty, it’s wild. That’s peak HN here

Very happy to see that I am not the only one. My pro subscription lasts maybe 30 minutes for the 5 hour limit. It is completely unusable and that's why I actually switched to OpenCode + GLM 4.7 for my personal projects and. It's not as clever as Opus 4.5 but it often gets the job done anyway

https://pdfling.com

A service to send pdfs to your kindle device through a chrome extension or web app.

Really enjoyed your talk two years ago :)

Add healthcare. Cannot send our patients data to a cloud provider

Yes, good point. Although the traffic for these websites is so small, that I think I‘m good there for a long time.

I’m a doctor myself and thus have contacts to people having medical practices :)

But at least in Germany there are some agencies doing nothing else.

I recently build a website for a medical practice using Astro.

I was amazed by how easy it was compared to my experience with Wordpress for this several years ago.

And I can host it for free on something like Netlify and I don’t need to worry about the site being hacked, like with WP.

I even built a very simple git-based CMS so that the client can update the content themselves.

Web dev has really come a long way, despite what a lot of people say.

Actually the abstractions are much thinner than with something like NextJS imo. It all comes down to what you are comfortable with. If you learned web dev in the React era, this approach feels very odd, but if you come from something like Ruby on Rails, this actually quite intuitive and not a lot of abstraction (see Jeremy‘s comment in this thread).

I personally like to stay with normal HTML and FrankenUI instead FastHTML instead of MonsterUI tho.

For the creation part of a KG I do understand this. But for inference and knowledge organisation, there is still value in graph based semantic structures imo

There is no compelling evidence that MRI contrast agent causes cancer. Gadolinium (the stuff that’s in the contrast agent) can deposit in the body, e.g. in the brain, but if this even has any consequences is still unclear. Nonetheless there is some nice research going on how to drastically reduce the amount of contrast agent needs to be administered through image postprocessing.

The name checks out for a sandbox! Well done

The same symptoms can be caused by either a blood clot OR by hemorrhage in the brain. If you give a patient with a hemorrhage fibrinolytics, you killed him. That’s why you need the CT first: to rule out bleeding.

Came here to say this

Sometimes people just like to build stuff for the sake of it.

This.

Just had a longer discussion with my wife just now why this is.

So many discussions in Germany revolve around ideology and not what’s best in the situation right now.

First, just because an example app stores the password in plain text, doesn’t mean you have to do the same. Hashing a password isn’t really that complicated.

I wrote on how to implement a simple login system in FastHTML [1]

Second, you can use fast tags in other python projects. Import it from fastcore and call the „to_xml()“ function on the fast tags. This will convert it to HTML

[1] https://blog.mariusvach.com/posts/login-fasthtml

Yes please. We hope to move away from our RIS and integrate our reporting workflow into our PACS this year

To be honest, I heard of several radiology practices that hand the patients a normal report directly after the exam and they look at the actual images only after the patient has left.

I guess the reasoning is that they want to provide „good service“ by giving the patient something to work with directly after the exam and the workload is so high that they couldn’t look at the images so fast. And they accept the risk that some people are getting angry because their exam wasn’t normal in the end.

But on the scale a typical radiology practice operates today, the few patients who don’t have a normal exam don’t matter (the number of normal exams in an outpatient setting is quite high).

I find it highly unethical, but some radiologists are a little bit more ethically relaxed I guess.

What I want to say is that it might be more of a structural/organisational problem than incompetence by the radiologist in your case.

(Disclaimer: I’m a radiologist myself)