I have a hunch about MRI that comes from seeing a company that an old uni friend has built around liver disease.
The original insight of my friend, who is a doctor, is to do with a particular liver condition which is (or was) diagnosed with a biopsy. Obviously that’s a medical procedure. It’s invasive. He discovered that the biopsy could be substituted with an MRI scan plus new techniques in computer vision.
And I wonder how many medical diagnoses are tractable to that same approach?
Add to this three points:
- MRI machines and the associated setup are pretty portable. I’ve seen pictures of an MRI unit in a shipping container, dropped into a hospital car park. So treat is like a black box: patient in; images out. Like any technology, the more MRIs that are run, the cheaper it’ll get.
- MRIs aren’t x-rays. As far as I know, you could have an MRI every day of the week and it wouldn’t do you any harm. You might get a bit of a headache from the clanking of the electromagnets as they quench but that’s it. (It was pretty noisy when I got an MRI for my knee, but I remember thinking at the time: I’ve paid more money for worse gigs.)
- Beyond the initial imaging, all the work of diagnosis is data and software. It’s quick, parallelisable, and machine learning models can be upgraded over time: hetting a second opinion is not like having to get back under anaesthetic to collect another sample for biopsy. As there is more training data and the software improves, second and third and fourth opinions can be run on the original images without having to return to the patient.
Put all of this together:
Could our future include pro-active regular screening for all kinds of conditions?
Imagine you get a full-body MRI every 6 months. Nothing wrong necessarily, it’s just like going to the dental hygienist. Then 100s of different machine learning models run, one looking for a particular liver condition, one looking at another organ, another looking for such-and-such anomaly elsewhere, etc. It’s purely precautionary; a way to pick up issues before they get serious; a Check Engine light for your body. You’d get a notification on your phone the next morning.
An app ecosystem around regular, precautionary MRI:
It’s unlikely that the company which is good at MRI machine manufacture is the same as the company which is good at customer relationship and operations, and it’s unlikely that either of those will have the software focus to train machine learning models to identify specific conditions (each condition probably being the topic of a whole stack of doctoral theses).
So I see something that is more like a software ecosystem. As a consumer, you pay (or your insurance pays) for the twice yearly scan. A portion of that fee gets divided amongst the hundreds of separate companies that provide computer vision modules that run across your full-body image, like paying for Spotify streams.
OR BETTER, to complete the feedback loop, each company might run their software on the image at their own cost, and they receive a success fee for each condition that they identify which is also successfully confirmed. (With some kind of adjustment to incentivise a low number of false negatives too.)
The role of the operations company is to orchestrate the ecosystem and economics, also managing the distribution to the computer vision app developers of training data (source imagery and eventual known outcomes from existing biopsy techniques and medical records).
One analogy is the company Planet which uses 200 satellites to take daily, high resolution images of the surface of the entire globe. Some of the satellites are to 50cm resolution.
Think of what you can do with high frequency global photography: look at infra-red signatures to figure out, to the day, when your crops are ready for harvest; monitor container ship positions to predict future pricing; see where to target your roof insulation sales effort; check for broken street lamps against a “known good” list; or whatever.
Sure you could achieve these by installing sensors, or using drones, or even walking the streets… but why bother? The satellite imagery has already been collected. The rest is software that runs on a schedule in the cloud.
Computers don’t get bored. Software is perfect for trivial or speculative repetitive tasks. It’s pixels and algorithms and compute cycles, that’s all.
I wonder what it would take to develop the technology (and establish the market, private or public) for this kind of MRI-based early warning system for personal healthcare. I can’t imagine that existing MRI tech would be a good fit for regular full-body – but maybe, one day, given the right incentives?
From a policy perspective: what kind of white paper would you have to write such that politicians would choose to fund this as industry-sector-creating R&D?
Anyway it feels kind of inevitable, though I’m not sure how we get there.
I have a hunch about MRI that comes from seeing a company that an old uni friend has built around liver disease.
The original insight of my friend, who is a doctor, is to do with a particular liver condition which is (or was) diagnosed with a biopsy. Obviously that’s a medical procedure. It’s invasive. He discovered that the biopsy could be substituted with an MRI scan plus new techniques in computer vision.
And I wonder how many medical diagnoses are tractable to that same approach?
Add to this three points:
Put all of this together:
Could our future include pro-active regular screening for all kinds of conditions?
Imagine you get a full-body MRI every 6 months. Nothing wrong necessarily, it’s just like going to the dental hygienist. Then 100s of different machine learning models run, one looking for a particular liver condition, one looking at another organ, another looking for such-and-such anomaly elsewhere, etc. It’s purely precautionary; a way to pick up issues before they get serious; a Check Engine light for your body. You’d get a notification on your phone the next morning.
An app ecosystem around regular, precautionary MRI:
It’s unlikely that the company which is good at MRI machine manufacture is the same as the company which is good at customer relationship and operations, and it’s unlikely that either of those will have the software focus to train machine learning models to identify specific conditions (each condition probably being the topic of a whole stack of doctoral theses).
So I see something that is more like a software ecosystem. As a consumer, you pay (or your insurance pays) for the twice yearly scan. A portion of that fee gets divided amongst the hundreds of separate companies that provide computer vision modules that run across your full-body image, like paying for Spotify streams.
OR BETTER, to complete the feedback loop, each company might run their software on the image at their own cost, and they receive a success fee for each condition that they identify which is also successfully confirmed. (With some kind of adjustment to incentivise a low number of false negatives too.)
The role of the operations company is to orchestrate the ecosystem and economics, also managing the distribution to the computer vision app developers of training data (source imagery and eventual known outcomes from existing biopsy techniques and medical records).
One analogy is the company Planet which uses 200 satellites to take daily, high resolution images of the surface of the entire globe. Some of the satellites are to 50cm resolution.
Think of what you can do with high frequency global photography: look at infra-red signatures to figure out, to the day, when your crops are ready for harvest; monitor container ship positions to predict future pricing; see where to target your roof insulation sales effort; check for broken street lamps against a “known good” list; or whatever.
Sure you could achieve these by installing sensors, or using drones, or even walking the streets… but why bother? The satellite imagery has already been collected. The rest is software that runs on a schedule in the cloud.
Computers don’t get bored. Software is perfect for trivial or speculative repetitive tasks. It’s pixels and algorithms and compute cycles, that’s all.
I wonder what it would take to develop the technology (and establish the market, private or public) for this kind of MRI-based early warning system for personal healthcare. I can’t imagine that existing MRI tech would be a good fit for regular full-body – but maybe, one day, given the right incentives?
From a policy perspective: what kind of white paper would you have to write such that politicians would choose to fund this as industry-sector-creating R&D?
Anyway it feels kind of inevitable, though I’m not sure how we get there.