I can understand why, when you're dealing with human lives, you wouldn't want anything but the best. But that wasn't the use case of the original article.
Also, I'm curious if good software can somehow compensate for low-quality hardware. The availability of a cheap machine could spur the development of new signal processing algorithms that work around hardware shortcomings.
I don't like this post because it just plays up the difficulty of the engineering without any context. Yes, making an ultrasound is hard and complex, but its not as hard and complex as making a car, which is how much they cost. He just presents the engineering like we should be able to intuitively relate that to cost.
Hell, he also doesn't contextualize any of the engineering itself. At the end he brings up the housing, and not making it cut or burn people. What? Not making things sharp is not rocket science. The machine doesn't cost 30k-150k because it isn't sharp. Most things aren't sharp. He's conflating details with challenges.
Acoustic lenses, 100 micron precision? Injection molding produces surface finish and repeatability to sub-micron levels without even trying. A 50 cent plastic magnifying lens is more precise than the acoustic lens in an ultrasonic machine, AND its made of more expensive plastic (it just so happens the best acoustic matches between water and PZT are very cheap plastics).
He's completely dismissive of PZT buzzers, which is really unwarranted. A 2 cent PZT buzzer is actually the closest product to an ultrasound probe, engineering wise. The transverse mode of a PZT buzzer is in the tens of MHz, the surface finish is easily a few microns, and the performance across the disk is very uniform- it just so happens the cheapest way to mass produce buzzers creates those properties. Of course to get two disks that are similar you need to pay more, but the price is still measured in cents.
Dice and fill is complex but not necessary, especially for lower frequencies. It uses standard semiconductor tooling and while expensive, it doesn't mean each transducer costs as much as a car. Splitting the elements is also complex, but its a hell of a lot simpler. Also he keeps saying 100 micron- its not a feature size except for dice and fill. It's an accuracy, and one that can be achieved by hand with 2x magnifiction. For actual context: 127 microns is your standard "sloppy" machinery tolerance. When you order something from a machine shop and mark a dimension as not important, you'll get 127 microns/.005".
There are any number of ways to section the transducer. Laser would be one. It would cost less than a dollar per, and worst case you have to repole the PZT afterwards. Bonding is a pain but not complicated. You can just solder to silver coated PZT, strain relieve the bond wire, and pot the whole thing in plastic. Keeping that 100 micron accuracy just means making sure you don't get hair all over it when you put it together, unless you're blond in which case your hair is too fine to matter.
And then... a cable? That's his next example of complexity? Buy a bunch of ground plane ribbon cables and roll em up. Crosstalk will be less than on a PCB. This whole thing just gives off the impression he's stuck thinking about things the way he's always done it and has never considered alternatives. He talks about the challenges of building products.
>Ever made a protoype? Ever made 2 of them? 4? 100? 10,000? And made them all the same? Yep, it's a different world when you have to start selling and supporting products.
It's one thing to make a product that doesn't inject noise into the hospital grid, is shielded to more than FCC standards, its waterproof and heat and fireproof. It's quite another to just make a product. Yeah, making things is hard, but literally every company that sells something knows how to do it. It's built into cost. It's not an explanation.
A transducer does not cost as much as a car (well, not a well made car), but a system does. That's the cart plus a number of transducers.
Yes, I only covered the transducer in that post, but as I noted, the systems and software deserve their own Pt II. It takes time to write this kind of stuff you know?
Your other points I try to answer in my updated blog post along with many others here in this thread.
Soldering with silver-lead depoles PZT. Bismuth solder does not and regardless repoling PZT is trivial- a couple hundred volts at most, and temperatures below the melting point of plastics.
The transducer is the only part of the system that should concievably be expensive, though. Doctors don't use special, medical-company made laptops- they use thinkpads or dell or whatever. The cart is unnecessary! It shouldn't be where the cost is coming from. The cart should add a few thousand dollars. The electronics have no business costing tens of thousands of dollars. The R&D, the software, and the transducers are the only plausible money sinks. If making a monitor is expensive, the companies should buy a cheap laptop and write an application to run on it.
I do really enjoy your posts but I am firmly in the camp that the price of ultrasound machines is an order of magnitude too high. I also think it needs to come down ASAP and that the future of cancer treatment really depends on regular ultrasound screenings and machine learning.
Edit: also, to back up my statements about the electronics: https://news.ycombinator.com/item?id=13245998