”Determining whether something is (non) malignant isn't trivial. It often requires a more invasive test (e.g., biopsy, exploratory surgery)”
Not necessarily, waiting can in many cases tell you whether something is malignant. If we had, say, a century of monthly full-body MRIs of a million persons, together with their history and cause of death, the technology to align those scans across time, and the technology to analyze such a data set, a retrospective cohort study (https://en.wikipedia.org/wiki/Retrospective_cohort_study) probably could uncover quite some interesting and useful information.
If, at every step, you pick the option with the highest expected QALYs (https://en.wikipedia.org/wiki/Quality-adjusted_life_year), and you manage to make doing the MRI very efficient, I think doing that experiment even might pass an ethics committee (at the cost of making the analysis harder)
(might because one could argue that the patient would be better of if the money spent doing those MRIs were spent on something else)
Unfortunately, doing such an experiment isn’t practical (¿yet?).
First, the cost would be insane. A century of monthly MRIs is 1200 scans per person, or 1.2B scans to finish your hypothetical dataset. We pay about $US 550/hr for scanner time, using a research scanner that's subsidized (i.e., we're just covering costs, not making a profit). The article doesn't say how long the scan is. You can burn as much scanner time as you want chasing resolution/quality, but an hour seems reasonable. That comes out to $660B, which is...a lot, and we haven't even paid staff yet!
Second, we have sorta tried this already. There's a massive neuroscience project to scan tons of brains called the Human Connectome Project. They have 1200 subjects, some scanned multiple times, behavioral measures, health outcomes, the works....
People have certainly found stuff in the data (I'm using some of it right now), but it hasn't lead to wild breakthroughs. There's a ongoing debate about whether this money would have been better spent on hypothesis-driven research instead.
For comparison, the entire NSF budget for next year is around $7B. That would just about cover the imaging component of building a speculative and ethically-questionable data set, assuming someone else pays for the biopsies, analysis, and staff.
(The NIH does have more money, but also funds trials, vaccines, and other stuff that we probably don't want to cancel for a century).
ML & better imaging algorithms should help a lot. Times could be reduced from 45min to 15min [1]. Scanning a beating heart can be reduced from 4min+ to 25sec [2]. Also ML can aid in comparing past images to current ones [3], which would give you quick insights into what changed.
Two ideas for time reduction I haven't seen discussed but perhaps might also help:
1. Don't scan at the same resolution across the whole body each time. Instead, focus on anomalous places that you wanted to monitor from the first high-res scan, or places that look anomalous in low res in the latest scan. Then dial up the resolution in those areas.
2. If better imaging algorithms existed that could account for very slight movements of the body (ala the heartbeat one above), perhaps prep time could be reduced by changing the physical layout of the scanner itself. The whole process of lying down and getting your head or whatever mounted in their plastic frame, then lifting the gurney and slowly sliding it into the machine.. it's all very slow.
Instead of laying down, what if you could just walk in and be upright and get scanned relatively quickly - basically a slightly longer but similar experience to a chest X-ray. There's upright MRIs right now but they're not very high res (0.7tesla instead of 1-3), very few exist, and I'm sure they still take some time to complete scans and have lots of error correction extra scans to correct for patient movement.
However, even with less time in scanner, there's a lot of fixed time cost of scheduling and patient prep (remove all metal things, wear this gown and these ear plugs, please don't freak out its really claustrophobic patient messaging), as well as still needing technicians.
I'm just a patient who's gotten some MRIs but it definitely feels like there's ways to cut the time down significantly.
Your first idea is pretty common. There's usually a very coarse "localizer" scan at the beginning of a session, which is used to set the field of view for subsequent runs. The whole body scans are (at least in theory) meant to find tiny things that are asymptotic though, so I'm not sure that searching through (say) 10mm slabs will help much.
Open, upright scanners do exist, but they're lousy. The goal of the (big) magnet is to produce an incredibly strong, uniform magnetic field. Due to physics, this is much, much easier to do with a torus-shaped magnet than a 'U'shaped one. Even so, there's one point ('isocenter') where the magnetic field is maximally flat and the quality will be best. The gurney moves to point the region of interest (e.g., your head) right to the isocenter. That's why the tech often uses a little light or laser to find your position, rather than just asking you to scooch. Motion is also, as you alluded, a huge problem.
I hate to be discouraging, but I am excited to see people actually thinking about MRI on HN!.
MRI is all about the protons. Under normal conditions, the protons in your body are all spinning ('precessing') around their own axis, but they're disorganized: each proton's axis is pointing in a different direction. They're like little wobbly tops drifting through space.
In an MRI scanner, the strong static magnetic field (B_0) forces the protons into alignment, so that their rotational axes are now lined up with the field's north-south axis. The field needs to be very strong for this to work, which is why MRI systems usually have expensive superconducting magnets.
Now that we've created a nice organized system, we're going to destroy it. A quick radio frequency burst energizes the protons and 'knocks them over' so they're no longer aligned with the field. Once the pulse ends, they 'relax' and realign themselves with the magnetic field, releasing some of that energy as they do so.
Sensitive detectors around the subjects' head detect those emissions and use it to determine how long it took for protons to realign themselves with the different components of the magnetic field. T1 is the time (or formally, the time constant) needed for relaxation parallel to the static field; T2 is the time needed for protons to relax to the transverse component. The T1 relaxation time for fluids is on the order of seconds, while fatty issue is more like 50-150ms. In the brain, grey matter has a relaxation time of 1.3 sec, but the fat-coated white matter relaxes much faster (~0.8), which makes T1 images very useful for examining brain anatomy.
Hopefully, this little crash course has revealed one of the bottlenecks in MRI: the actual signal being measured is slow.
Of course, I've massively oversimplified things and didn't explain at all how we localize these responses. The proton's precession frequency depends on the magnetic field, so by changing the static field slightly (across space), we can measure T1 at different locations, and sometimes even overlap measurements. You can't switch the field too fast though, or you'll start to induce currents in the subjects' nerves, which hurts. This is actually the principle behind a brain stimulation technique called transcranial magnetic stimulation.
There is still tons of room for improvement. Stronger fields lower the relaxation time, so the scans are faster (and the relaxation time estimates are better). Improvements in the RF coils help a lot too: the signal being measured is very faint and there's a lot of self-cancellation. On the software side of things, a lot of effort has already gone into designing clever pulse sequences—and the sophisticated signal processing needed to interpret their results.
Things will obviously continue to get better; I was just looking at some data from ten years ago and it looks awful compared to more recent scans.
That said, the "just use machine learning!!!" tone in some of the comments is kinda frustrating. Most of the people in this field aren't dummies--if it were as easy as downloading PyTorch, someone would have done it already. It turns out that the biology and physics are both stupendously complicated (and fascinating too).
“Because it’s too expensive,” seems like a great opportunity for some clever startup to figure out a way to make it less expensive. Computers used to take up entire rooms. Flying across the country used to be insanely expensive. Cell phones used to cost a ton of money per minute. Reducing the cost of scanning, or developing entirely new scanning tech isn’t science fiction, it’s the future.
The doctor gets $98. The $550/hr price I quoted is for a research scanner where no one is even trying to turn a profit; they just want to pay off the machine and its operating costs.
The machines themselves are not magically cheaper in Japan; they're just being paid for through some other route. If the scanner were somehow free (gov't grant?), 98$ seems pretty reasonable for the labor.
Again, that’s what you pay (near as I can tell), but that’s not what it costs.
They’re also “free” (at the point of service) in the UK, but the scanner and helium are not a generous gift from the fey folk, nor does the Queen volunteer as a tech.
These things cost money. Any study you do is going to have to cover expenses—-including their share of a temperamental, multimillion dollar machine (or find a way to dip into the same accounts that cover its clinical use).
A medical grade MRI scanner costs around $1 million, and maintenance costs are circa $100,000/year. With good utilisation, say 3000 hrs/year, the machine costs might be around $70/hr.
The $550/hr you are paying may be because you are using a particularly fancy scanner, or because utilisation rate is low, or because the amount is loaded with overheads, or w/e.
That’s in the ballpark for a 1.5T, but those are fairly old. About $1M per Tesla used to be a decent rule of thumb, but it’s come down a little at the low end. Still, I would be amazed to see a 3T for anything below $2M.
As for the $550/hr, it’s probably true that research scanners have lower utilization and higher costs to support all the weird stuff researchers want to do. An outpatient clinic specializing in knees can run much leaner. That said, that rate seems to be pretty standard across universities and I maintain that it's a very reasonable estimate of the cost. For example:
Most of these do not include F&A. It’s already coming out of the grants and external users pay a "Dean's Tax" on top of that to cover the missing overhead (which can often double the price).
Not necessarily, waiting can in many cases tell you whether something is malignant. If we had, say, a century of monthly full-body MRIs of a million persons, together with their history and cause of death, the technology to align those scans across time, and the technology to analyze such a data set, a retrospective cohort study (https://en.wikipedia.org/wiki/Retrospective_cohort_study) probably could uncover quite some interesting and useful information.
If, at every step, you pick the option with the highest expected QALYs (https://en.wikipedia.org/wiki/Quality-adjusted_life_year), and you manage to make doing the MRI very efficient, I think doing that experiment even might pass an ethics committee (at the cost of making the analysis harder)
(might because one could argue that the patient would be better of if the money spent doing those MRIs were spent on something else)
Unfortunately, doing such an experiment isn’t practical (¿yet?).