Comment by tacon | Hacker News Reader

Comment by tacon | Hacker News Reader

Parent Full thread

A few years ago, my patellar tendon was getting sore from running, and I went to a sports medicine specialist. At that first visit, he did an ultrasound scan over the length of my patellar tendons, and it was very discouraging to see these "black" sections in the mid-core of my tendons. They were actually tissue that was a bit disrupted, but the way it scattered the sound, it appeared as "black". Kind of freaked me out that my tendon was "dead" in the middle. The actual diagnosis was patellar tendinopathy, previously referred to as patellar tendinitis, or "jumper's knee".

He gave me exercises to build up my tendons, mainly eccentric resistance exercise, but it was going to take a long time. I wondered how I could monitor the progress of the healing. The doctor visit and scan was at least $200/visit, and the healing could take a year. On AliBaba, I checked how much a low end machine would cost to use at home.

There were all sorts of machines available, and one manufacturer's rep latched onto me. I still get the occasional email checking if I'm ready to buy. I think these machines may require some doctor's approval, too, but I never went that far.

But they sure don't cost anything like the prices in this article. In 2013, for quantity 1, IIRC, I was quoted $1100/each for this SUN-806F laptop plus sensor plus software unit:

http://www.cnsunbright.com/china/english/newshow.asp?Cid=108...

Maybe I'm missing something critical in this price difference? As the OP was explaining, you can almost build one yourself.

OP/author here, I also have patella tendinitis since I dislocated my patella when I was 17.

This is the reason I started looking at ultrasounds, I figured that I could buy one for the $160 cost to get a scan from a doctor. I have looked around for a cheap ultrasound, but $1100 is far cheaper than I was able to find (I got to about $3k but looked like a dodgy dealer so didn't link to it). But 1k is still way more expensive than I think it should be.

It should be a sensor you just plug into and iphone for a few hundred. I mean there is the http://www.thermal.com/thermal-cameras/ thermal camera for only $250, why not ultrasound? Making a photo sensor must be thousands of times more complicated than ultrasounds sensors.

I just want to know why I can't buy one cheaper than my massively more complex iphone :)

With the math on transducer costs, you've basically answered the question. However, aliexpress will get you 2MHz transducers 10x cheaper if you buy them in bulk. Then of course you have to worry about dodgy chinese website component quality.

I also think you're seriously underestimating the computational cost of converting the measured signal to a usable image. No way is an RPi fast enough to produce a live image with decent quality. There are quite a few papers in the literature about using high-end GPUs to speed up ultrasound imaging and denoising, for commercial vendors I'm guessing it's custom ASIC territory.

> I also think you're seriously underestimating the computational cost of converting the measured signal to a usable image. No way is an RPi fast enough to produce a live image with decent quality. There are quite a few papers in the literature about using high-end GPUs to speed up ultrasound imaging and denoising, for commercial vendors I'm guessing it's custom ASIC territory.

Really? Medical ultrasound has a long history predating high-end GPUs. Maybe that kind of computing power is required for modern high-resolution/low-lag devices, but my guess is you could get something usable with a computer as powerful as a RPi.

http://www.ob-ultrasound.net/history1.html

I can confirm that you can turn ultrasound data into a 2D image using a simple mid 2000's era ARM processor and an FPGA. So a RPi can even be overkill for that purpose.

But basic ultrasound machines have been with us for decades. From a quick check, since the sixties! I have no hard data yet, but I would be surprised if a Pi couldn't do it.

3D scans are another matter, but this is a thread about low cost scanners, so that's out of the picture.

>Making a photo sensor must be thousands of times more complicated than ultrasounds sensors.

Cheaper thermal imaging sensors like that usually use a very, very low resolution sensor for picking up thermal data, then use interpolation to map the low-resolution data onto a high-resolution image provided by a higher-res conventional camera sensor. Obviously, this approach won't work with ultrasound tech, as there wouldn't be a secondary high-resolution data source.

In the post there is a link to a paper where someone used a 40x40 array for 3d ultrasound imaging and there is a 64x array at 35MHz used for high resolution imaging of things like eyes. That is pretty low resolution for such interesting applications. Imagine the applications for a high resolution 3D ultrasound scanner combined with a high-resolution camera.

A 40x40 ultrasound array can scan with a much better resolution, look at prenatal imaging for example. The phased array is modulated in the time and space domains, so one cell can substitute many fixed ones. With the speed of sound, the receivers resolution of depth information extracted from a reflection's timing information would be a million times better. ...

I'm not sure I would understand my own description, let me go again: The transducer sends a modulated signal and receives the reflections. From the time difference information is gained. The signal is steered using beamforming as known from radar technology. This creates many different impulses that can be correlated from the received reflections.

The key is that the 1600 sells deliver more than one input over time, from a nonstatic signal, which is like shooting photos from different perspectives. I believe cell correlation or whatchamacallit is done in all kinds of variation for CCD data, too.

It might be helpful to elaborate on what applications a 3d ultrasound scanner would have.

> Making a photo sensor must be thousands of times more complicated than ultrasounds sensors.

What makes you think that?

Photo sensors have no moving parts, photons are the carriers of the electromagnetic force, so they readily interface with electronics given the right interface material; any semiconductor material with a matching bandgap will do, photons interacting with a bandgap produces free charge carriers; this is exactly what you're after. The rest is chipscale manufacturing, i.e. lots of sensor cells in a repetetive pattern. You can easily mass produce that in large quantities using photolithography methods, yet even if your process were to bad, that you'd get a low yield it would still be profitable.

Ultrasound on the other hand is acoustics, i.e. mechanical waves. So you're dealing with moving parts, mechanics. Transducers are discrete parts, they can't (yet) be manufactured at chip scale. They're ceramics, and there are lot of parameters that go into manufacturing ceramics which limit the amount that can be produced in a single batch. And then you've to deal with electronics that's not over waaay beyond your typical hobbist level; it's in fact so complex that it goes over the head of most professional electrical engineers and there are companies who are specialised in ultrasonics transducer electronics consulting, doing the difficult design work for the large brand names.

part of the reason thermal cameras are cheap and ultrasounds are not is that thermal cameras have consumer uses. One can use thermal cameras for hunting or determining home energy efficiency, an ultrasound can pretty much only be used for medicine. Although, consumer sonar for fishing(which operates at lower frequencies) is still fairly expensive[0].

Second, your estimate on the price of PZT elements is not necessarily realistic as the transducer is rather large(22 mm in diameter). A 10X10 array would be impractically large. One can buy PCB mount piezo elements for motors for cents per element in bulk[1]. It might even be possible to modify one of those expensive components into an array as was done in this homemade STM[2]. Although in both cases there are bound to be issues with impedance matching and potentially changes in resonant frequency.

[0] http://www.westmarine.com/sonar [1] http://pcbmotor.com/applications-and-solutions/smd-piezo-com... [2] http://dberard.com/home-built-stm/

Used ultrasounds can be had for good prices: https://www.dotmed.com/equipment/2/4/all/

> I just want to know why I can't buy one cheaper than my massively more complex iphone :)

Size of market for iPhones compared to size of market for iPhone ultrasound devices. Also, you'll need FDA approval in the US for their use.

You would also need approval in the EU and in other countries around the world. There are many standards that medical devices must meet. It takes a pretty good sized company to do all things that must be done in order to actually design, manufacture, ship, surveil, maintain and end-of-life a medical device of any complexity.

Size of market is a chicken and egg problem, if they were cheaper, the market could grow, and get more cheaper. Also FDA would only regulate for medical use, it could be sold for educational use to remove the regulations.

With respect, I think you're speculating here.

For example, the FDA and other agencies are concerned with the effects of transmitting energy into the human body. Medical devices like ultrasound systems are not toys.

I don't mean to be harsh at all, but with respect, I question your assertions here.

It would be interesting to see, I mean the argument is that my head phones don't need FDA approval and they can damage my hearing if I turn them up too loud. But some kind of limits would be important to ensure safety.

And this post is a question because I don't know the answer, so I am just guessing (a lot). I hope someone with actual experience in getting FDA approval for a medical ultrasound machine actually reads it and answers, with how much it cost and what hoops they had to jump through.

I have worked in the medical device industry building ultrasound systems for twenty years.

In general, there are thousands of regulations that must be satisfied. Some specifications that must be met include ISO 13485, ISO 14971, IEC 60601 3rd Edition, IEC 62304, and probably ten more that I have forgotten about, such as RoHS, WEE, radiated emissions, etc.

The systems I worked with involved multilayer (16 layers plus) circuit boards, custom ASICs, FPGAs, ARM SOCs, etc., mixed OS (e.g., Windows CE and WindRiver, etc.)

The code base is large and complex, as you have not just ultrasound, but often, calculation packages of various kinds (cardiac, OB, etc.) These packages are expensive to develop and have to be carefully verified. If you're measuring the length of a fetal femur and translating the measured length to an estimated gestational age, you don't wish to be wrong. Same deal with cardiac output measurements, etc.

Then there is the whole issue of transducers. It's a complex field, and you have to deal not only with materials (some of which contain RoHS regulated elements, such as lead), but also, dicing saws (to cut the piezo), matching layers, transmission lines (to send signals to / from the transducer and the ultrasound machine front end.)

I'm not trying to be rude, or baffle you with a lot of jargon, it's just that my experience is that building a useful ultrasound machine is much more complex than it might appear from the outside of the industry, so to speak.

Finally, I think building an iPhone is probably even more complex, but the incredible volume of sales tends to make amortizing the cost more palatable.

I could be wrong, of course, and your comments are welcome.

Hey Jes, would be glad to get your insights on echopen.org 's project still =)

Electronics wise, we are using a single element, mechanically translated (similar to the atl access pv10 probe) - then with a single element you can cut on the price (so far our prototype parts costs 250$/300$) while still giving a basic image for the user, who doesn't all the time need all the feature of a high-end ultrasound scanner =)

What do you think?

That is a great post. I quoted it in the post if that is cool and will read up on some those references.

Do you think that a lot of that complexity would be reduced by opening it up to a wider development community through open source? How much open source, consumer electronics, were used? Did you source the transducers from another company, or make them in house?

Cheers again for the comment.

Thank you for your kind words.

The post is necessarily incomplete; I don't think I can give you a full appreciation for all that is involved. And I don't wish to give the impression that ultrasound is necessarily more complex than other imaging modalities, such as MRI, PET, etc.

I wouldn't rule out open source, and many open source tools are used in the engineering process (e.g., gcc, git, etc.) I don't know about including open source components in a commercial code base, though. The lawyers usually ran away screaming whenever we would suggest using any open-source code.

Certainly the high-volume electronics industry makes components that are used in ultrasound systems. Some of the ARM SOCs, for example, and I suppose, some of the FPGAs, DSPs, etc. There are also a number of custom components, such as ASICs and combined analog-digital parts that show up in the so-called "front end."

The companies I have worked for designed their own transducers. That really is a science as well. Making a transducer probably involves at least 100 - 150 process steps, and some very expensive equipment (dicing saws, for example.) You also often have injection-molded parts (e.g., the transducer handle itself) that are expensive to produce (the molds for the injection molding machine are done with 3D solid modeling tools, then CNC machined from steel, etc.

You also have whole departments of people that work in "Regulatory Affairs" and/or "Compliance Engineering."

You also need to employ people to "optimize" your ultrasound images, in each of the different modes you might support: B-mode (2D), doppler, M-mode, color power doppler, etc. You also have things like harmonic imaging, etc.

Then there is sending images you have collected off to the various PACS systems that hospitals have. This usually involves a complex protocol known as DICOM.

Ok, I'm out of gas for this comment. Again, not trying to drown you in jargon, just suggesting that there is a reason that new ultrasound machines tend to be pricey; that reason is that NRE (non-recurring engineering) is very expensive, actually building the boards is expensive, shipping the systems and maintaining them in the field is expensive, and maintaining all of the other systems you have to have in place to be a viable ultrasound company is really expensive.

Hope this helps; thanks for letting me share a bit of my experience.

On the topic of Open Source, our IP attorney was primarily concerned with licensing, but that's because he's an IP lawyer. Our Regulatory and Quality Assurance people would be far, far more concerned with the Design History of any open source code we tried to bring into a medical device product. How was it developed? Where are the requirements? How was the output validated against the requirements? Our process mandates Design and Code inspections, can we show that they were performed on this Open Source product, etc, etc, etc.

Once you see the uphill road, you realize you're better off starting from scratch and writing the product yourself instead of depending on OSS. For some noncritical aspects, it's fine, but then you get into the definition of "noncritical" :-)

That's an excellent observation. IEC 62304 speaks to these issues, although not in the context of open source per se.

It also depends on the class of your device (A, B or C in 62304 parlance) how much rigor you must put into your software process.

One more warning... Two years ago I was on a project that spec'd inexpensive transducers from China. We bought samples from multiple factories and they were all terrible. Datasheets were vague, and a lot of the transducers didn't even meet those weak specs.

Ultimately it didn't matter much for our beauty product but it would be a big problem if we tried to use them for imaging.

Yup. Yet one more cost: Supply Chain control is vital.

Again, valid point.

At one point in time, I had a number of people in incoming receiving working for me. We had to do all kinds of quality control on incoming components, such as castings, PCBs, etc. We had automated measuring machines, etc. It's all very expensive to set this stuff up and staff it.

We also had to use an XRF gun on parts, to make sure we weren't being shipped non-RoHS parts.

As I remember, one of the add-ins for our Agile PLM system (Agile is an Oracle product) was on the order of a few hundred thousand dollars for licensing fees and setup, not counting the months of consulting time necessary to load bills of materials, etc., into the system.

> You also often have injection-molded parts (e.g., the transducer handle itself) that are expensive to produce (the molds for the injection molding machine are done with 3D solid modeling tools, then CNC machined from steel, etc.

Just to point out injection molding, 3D solid modelling, and CNC machining don't have to be super expensive these days.

Large volumes, and especially high end stuff... sure that takes a lot of up-front cost. But for smaller volumes (eg prototyping, short run) the costs are very reasonable now.

It seems due to continuing development of Open Source hardware and software in these fields, which created good virtous cycles for them. Hopefully this continues... :)

Note - saying the above as someone who's into 3D solid modelling and CNC machining already. Haven't gotten into injection molding yet, but it's on my personal ToDo list for not far off. :D

Building anything is much more complex than it might appear from the outside. Yet, except on a few areas, people get ways to make stuff cheaply.

The fact that this company has to work with very low level components and that it had no spin-off on other areas is a market flaw, not something inherent on their devices.

If my previous comment is not enough detail I'll happily expand.

There are various ones for about 1k on ebay eg. http://www.ebay.com/itm/A-Portable-Digital-Ultrasound-Scanne...

Don't know if that'd the kind of thing or too basic?

So it was 3300 USD total for the solution?

And the interesting detail, in the article that the author links there is this opinion from the doctor who uses the portable ultrasound device:

http://www.kevinmd.com/blog/2014/05/pocket-ultrasound-machin...

"Other doctors often ask me if I bill for my exams. I don’t, because billing and the detailed documentation and posturing that would be necessary to prove to an insurance company that an ultrasound was necessary would take more time than I have."

The rest of the article is also insightful.

A few months ago I almost ended up with a functional one from a local maternity ward for less than $100 on publicsurplus.com. I didn't realize the thing weighed =~ 400lbs until after bidding on it. Thankfully somebody else got the auction at the last minute.

Reply on news.ycombinator.com