jmhmd

This is a critical point. I am curious what the team building this looks like? Do they have ultrasound physicists and clinical practitioners in addition to the AI researchers?

Yes, I was thinking about FUS as well! There are clearly ways to penetrate bone, but I have not seen it used for imaging, only for ablation. But I am not an expert there and it sounds like you have more knowledge in that area than I do.

Pedantry appreciated.

The machines are expensive (millions range per MRI scanner), staffing the machines nearly around the clock with highly educated technologists, repair/maintenance of expensive specialized machinery, radiologists to read each scan (esp with a current shortage), means it’s very expensive to set up and run an imaging center. Opening and owning an imaging center used to be seen as fairly lucrative. and many radiology private practices did just that, however, the economics have changed over the years with dropping reimbursements, staffing shortages, etc and now often these imaging centers are seen as a liability rather than an asset.

Given that current ultrasound probe technology (including butterfly) relies on the probe being essentially in contact with the tissue being imaged, it’s hard to imagine how this set up can be effective with the imaged volume so far from the transducers, since there will be a huge amount of dissipation in the water bath, but maybe they have found a way to solve that? Also, I imagine that the quality of the images, such as they are, will fall off very quickly in larger patients. Will be interesting to see.

Not taken negatively, and you're right. This is even more difficult with things that are opinion, and not clearly verifiable.

Hah, I'm neither a bot nor written with any help from an LLM, but I'll take the fact that you can't tell the difference as a compliment :)

That's because there isn't any data yet, at least not enough from real patients to be meaningful. I would love to see some of the raw imaging data they have generated though, if that's what you mean.

Your points are well taken, and I think this is the fundamental struggle of anyone who works in a narrow and deep field. It's truly difficult to see things from a different point of view sometimes. It certainly could turn out that this ultrasound setup gives truly new information, but, it isn't really a new way of generating an image, it's the same physics we've used to generate images from sound waves for decades, and that modality comes with some pretty hard physical limitations that this demo does not directly address. Time will tell, if they don't run out of money. I'm hopeful!

That's generally exactly what we do, which if we need to follow 2x or 10x incidental lesions in the population, leads to cost and availability problems. A lymphoma patient in remission needs follow up scans too, and I don't want them to have to wait 3 months because thousands of people are now following up their benign adrenal adenomas.

Some initial thoughts as a practicing radiologist:

- This looks really cool and I hope they keep innovating on this. I love seeing new modalities develop and despite my (many) reservations and criticisms, if even one good use case comes out of it that truly helps people, it's tech money well spent imo.

- They show the reconstructed images as though they are a low resolution CT, and promise that quality will improve as they iterate. This is cool, but ultrasound is not CT. Ultrasound cannot image the lungs, as they are filled with air. You cannot find bone lesions, as the sound waves do not penetrate the cortex. You cannot image many structures in the abdomen if they are surrounded by gas-filled bowel. The brain is encased in bone, so you might get some penetration but it will be very limited. Even with theoretically perfect AI reconstruction, these scans will not be true "full body" in that there will be structures that are not reliably imaged. Imagine paying for weekly full body scans for years, everything looks fine, then its the lung cancer surrounded by air and invisible to ultrasound that kills you (that's why we use CT for lung screening!)

- The images they show are very cool, and do appear to show the correct structures. I realize this is early, but fuzzy shapes of organs is very, very far from medically useful. The whole point of screening is to identify problems early, often by definition, small. This technology looks like it will be best for seeing large, superficial (close to the skin) structures, whereas for effective screening, you want the opposite - small, deep structures.

- "Incidentalomas" or unexpected, probably benign, findings are annoying to physicians, but I in general have no problem with people collecting data on themselves where they can. To me it's similar to heart rate monitors or home blood pressure cuffs. The main issue here is education, so that patients know what the data is and is not telling them. The more complex the data, the more difficult that is.

- Many people mistakenly believe that early diagnosis is the final boss in medicine, that if only we could find every cancer early we could prevent all those deaths. There are, in fact, many, many other hurdles and bottlenecks. Many chronic, expensive diseases do not have clear imaging manifestations. The claim that "it's completely possible that with enough early imaging in the future, the world could avoid 30% of all deaths and 50% of all healthcare costs", I think, to any practicing physician, would sound completely divorced from reality.

It depends on the indication for the scan. Some indications do not require contrast, others MUST have contrast in order to have any value. If you refuse contrast without understanding the reason, you may be simply wasting your time and money.

I think you must have misunderstood where the artifact was coming from. Gadolinium retention has been shown to occur, but has not been reliably linked to any clinical symptoms. Gadolinium tissue retention also does not interfere in interpretation.

I agree with this sentiment. I have always wished, maybe naively, for the type of computing environment that makes possible things you see in sci-fi movies and shows, where someone can simple "route all power to the forward lazers!" or "use the power cells from your rifle to keep life support systems online!" This imaginary world where technological components are trivially interchangeable, compatible, reusable. My impression is that if you even asked a smartphone hardware engineer to replace a broken iPhone camera with a leftover working camera from an Android phone that, at best it would be an extraordinarily difficult task, and at worst, just may not be possible.

The issue is that in medicine, much like automobiles, unexpected failure modes may be catastrophic to individual people. “Fixing” failure modes like the above comment is not difficult from a technical standpoint, that’s true, but you can only fix it once you’ve identified it, and at that point you may have a dead person/people. That’s why AI in medicine and self driving cars are so unlike AI for programming or writing and move comparatively at a snails pace.

Poorer performance in real hospital settings has more to do with the introduction of new/unexpected/poor quality data (i.e. real world data) that the model was not trained in or optimized for. They still do very well generally, but often do not hit equivalent performance to what is submitted to the FDA, or in marketing materials. This does not mean they aren’t useful.

Clinical AI also has to balance accuracy with workflow efficiency. It may be technically most accurate for a model to report every potential abnormality with associated level of certainty, but this may inundate the radiologist with spurious findings that must be reviewed and rejected, slowing her down without adding clinical value. More data is not always better.

In order for the model to have high enough certainty to get the right balance of sensitivity and specificity to be useful, many many examples are needed for training, and with some rarer entities, that is difficult. It also inherently reduces the value of the model it is only expected to identify its target disease 3 times/year.

That’s not to say advances in AI won’t overcome these problems, just that they haven’t, yet.

While a lot of this rings true, I think the analysis is skewed towards academic radiology. In private practice, everything is optimized for throughput, so the idea that most rads spend less than half of their time reading studies i think is probably way off.

As a radiologist and full stack engineer, I’m not particularly worried about the profession going away. Changing, yes, but not more so than other medical or non-medical careers.

litevna.app - a DICOMweb compatible medical imaging archive, built on cloudflare workers, to optimize fast image delivery globally. Images are all encoded as HTJ2K for progressive image loading, and the popular OHIF zero-footprint DICOM viewer is built in.

Building mainly to power the next generation of pacsbin.com, but may offer as a standalone service as well.

I use JWTs to let me do auth on cached resources. I can verify permissions in an edge worker and deliver the cached resource without needing to roundtrip to the database. Not sure how to implement that without JWT (or rolling my own solution). Lots of people here saying some version of “I don’t see the use case, just use X”, but these kinds of standards nearly always arise as a result of a valid use case, even if they aren’t as common.

Portable low field bedside MRI has been on the market in the US for a few years. See Hyperfine Swoop.

S3 is really cheap, and from the beginning I spent a lot of time optimizing for simplicity and efficiency, so I don’t have much overhead. Since this is targeted towards residents and academic radiologists, it’s important to me to keep the price as low as possible. What is your platform if you don’t mind sharing?

I started and run https://pacsbin.com, a radiology teaching file/research platform. I’m a radiologist and started this as a resident while unsatisfied with all existing options. It has been really gratifying to work on a niche problem for which a lot of my colleagues need a solution, and has helped me learn a ton about the tech and standards that underpin my profession.

It’s funny to me as a physician to see “you’re not a hospital” as an example of a system that cannot tolerate downtime. Epic, probably the biggest EHR provider in the US, has planned downtime for upgrades at least monthly, for 30-60 min each.

To be fair, this is not how doctors want to practice medicine.

You are partially correct, but I think the value is not in generating a draft for the radiologist to review, but to take a radiologists “raw” dictation, which pertains only to relevant findings in a case, and transform that into a complete, well formatted report. This is what scribes do in lots of different medical specialties, and save a ton of time. In my experience, reviewing pre drafted reports, as from residents, can often make my job slower rather than faster, as you have to read the whole thing to know what needs editing.

Also, at least in my practice, I would double the number of CTs read in an average day :)

The main thing that House MD has, that no other doctor in the world has, is not so much his superior intellect. It's that he and five other doctors spend 100% of their time on a single case, and can sit around all day discussing it, trying different things. If real world doctors had even a fraction of that luxury, you would see a lot more of what you describe.

There was this giant thing called healthcare right, and its main purpose is improving health—trillions of dollars are spent trying to do this.

As a physician, lol. Lmao

The BBB is a part of capillary walls and so there really isn’t a good way to “manually” inject beyond it. That said, there are current trials ongoing to temporarily disrupt the BBB using high intensity focused ultrasound, allowing injected chemotherapy to cross and help treat brain cancer.

No, identifying the tumor is only step 1, and is the easiest step. Most non-radiologists can identify whether a tumor is present. The harder part (and the true value of radiologist reads) is everything that comes after finding the tumor: what structures are the tumor invading? Is there spread to lymph nodes? Are there secondary findings that might affect the diagnosis or treatment?

These questions and their relevance changes for every individual case, and while each question by itself may be approachable with AI, getting a detailed and relevant report without meaningless noise from an AI ensemble is a very very hard problem.

CT is often done both as a non-contrast study (to exclude hemorrhage or large established stroke) and also as a CTA and sometimes with perfusion. This data at most stroke centers is all that is needed acutely to decide whether and how to treat, and MRI is often not done until days later to evaluate final extent of the stroke.