I've had this idea floating in my head for a while. Provide free annual full body MRIs to people, including being read by a radiologist, which they can take to anyone they want (doctor, lawyer, whatever). In exchange, they agree to send the org a copy of all their past and future medical records. The records are anonymized and tied to the MRIs.
The MRIs and medical data is then pumped through machine learning algorithms to attempt to identify early signs of disease. The data of course would be most useful to future people, because we won't know someone has something until a doctor diagnoses it, but if the AI model can look at the MRI from five years ago on everyone with disease X and find a common early warning sign, then we could apply it to future cases where it might be possible someone has X and catch X much earlier.
Basically, I want to create a charity that gives free full body MRIs in exchange for a bunch of medical data for training an AI.
So, these are not medically indicated scans which means you're going to be causing huge amounts of over-diagnosis and thus over-treatment, which cause harm.
Hypothetical Bob has his first scan, and it returns a few shadows on his lung. The doctors ask if he's ever had chicken pox and he doesn't think so but he's not sure. So, these shadows are probably just scaring from chicken pox. But maybe they're cancer. The only sure way to tell is for a biopsy. Every medical intervention carries risk, and because you've increased the numbers of interventions you've increased the numbers of people falling victim to that risk.
Bob got a scare; he got an unnecessary medical intervention; he was subjected to risk: what did Bob gain from this?
The only thing this extra data tells us is "chicken pox sometimes scars the lungs and they sometimes show up on MRI", which is something we already know.
You do have a good point about the difficulty in getting hold of data to train AI. Sadly, because of recent missteps in England that's going to continue to be difficult unless the company can show rigorous protections.
> So, these are not medically indicated scans which means you're going to be causing huge amounts of over-diagnosis and thus over-treatment, which cause harm.
What if we did something like set up a system where Bob's doctor can put in other symptoms, and only if one shows up that would indicate a lung MRI would the doctor be told, "Oh hey, we already did that, here's the MRI, and there was some shadow on the lung"?
> agree to send the org a copy of all their past and future medical records..
Even ignoring the issue of false positives, which alone is serious enough to make this a dubious idea... that is an extreme thing to ask of people. All my future records? I have no idea what my future holds, and what I may feel about that as I age and experience new problems. Nor how my family might feel about it on my deathbed.
I applaud new ways to for early detection of problems, but not like this.
Well if you didn't like the terms you wouldn't have to get your free scan. :) I hear where you're coming from, but the data isn't nearly as useful if we don't get to find out what diseases you get later in life. The whole point is to train an AI so we need the future data to make it work.
Except no profit would be made on data. All the findings would be given away for free. So we'd need donations to pay for the machine, its upkeep, and the staff to run it.
So not really anything like the pay with data model.
I've had the same idea for a long time, and I'd also like to see a charity start doing this. It's crazy that so many people are saying this is a bad idea. I think it has the potential to save millions of lives.
I was very tempted to join an AI startup in a similar space, but I decided to turn down their offer because they were a for-profit company, and that just didn't feel right. The offer was also too low, but I would have probably joined if they were a non-profit.
Not to be too snarky, but...you know about universities right?
I can almost guarantee there's a university or research institute near you doing some kind of research that's trying to improve or understand human health via MRI.
I actually never considered working with universities. I'm not too sure what that would look like.
My plan was to start a non-profit organization and hire some radiologists, AI experts, and software engineers. I would also need a lot of help from lawyers and healthcare experts to make sure we're complying with laws and regulations. Then I wanted to purchase a few second-hand MRI machines and start a pilot program somewhere.
I would probably start in New Zealand (my home country), and then move to other countries with universal healthcare (Australia, Canada, UK.) Early detection of cancer should end up saving a lot of money (compared to treating cancer in the later stages), so in the future it should be possible to get government funding that covers the entire program.
I'm not too interested in starting a for-profit company, but it's probably the only way to do something in the US. So I'd leave that to someone else.
But yes, it definitely makes sense to work with a university or hospital. Maybe the Centre for Advanced MRI at Auckland University [1].
(P.S. This is just a project I would want to tackle if I ever sell a company and end up with a ton of money. There's no chance that I'll be doing something soon.)
The thing you're describing is almost exactly a university medical center: a non-profit staffed with doctors and researchers running research projects. They also have the technical and administrative staff to run them properly.
Most professors at a medical school don't lecture very much, but are instead responsible for running research programs like the one you're describing. They either do this on their own, or as part of bigger consortiums. I'm mostly interested in brain stuff, where there are some big imaging+analysis collaborations. The Human Connectome Project collected about 1200 subjects' MRIs, plus a bunch of related health data. The Alzheimer's Disease Neuroimaging Initative is doing something similar, with a focus on neurodegeneration. I would imagine there are similar ones for the rest of the body.
Most people start on these projects as grad students, coming from biomedical programs (Neuro, Bio, Psych), engineering, physics, CS, or even math, depending on exactly what you want to do. There are other career tracks though--some people get hired on as research assistants or staff programmers, either attached to large programs or as permanent staff at big universities or institutes. These positions are a little harder to find, especially since RA can mean anything from "will train" to "needs PhD + 5 years experience". They often pay a bit better (though not tech industry levels), but have a bit less freedom.
The US actually funds a ton of biomedical research through the NIH (and NSF and DARPA), and is probably one of the better bets for running large, expensive studies.
The MRIs and medical data is then pumped through machine learning algorithms to attempt to identify early signs of disease. The data of course would be most useful to future people, because we won't know someone has something until a doctor diagnoses it, but if the AI model can look at the MRI from five years ago on everyone with disease X and find a common early warning sign, then we could apply it to future cases where it might be possible someone has X and catch X much earlier.
Basically, I want to create a charity that gives free full body MRIs in exchange for a bunch of medical data for training an AI.