Cracking the puzzle of automated low-cost malaria diagnosis | Hacker News Reader

Cracking the puzzle of automated low-cost malaria diagnosis

twitter.com

122 pointsDowwie32 comments

I just did a quick visual scan of the PDF. Some parts have to be fabricated. Custom parts of the microscope were designed with Autodesk Inventor Professional and 451 fabricated by Protolabs and 3D Hubs... I do not think this is open source. At least I do not find a link to the file.

The measurement results were 465 saved as CSV files and processed with MATLAB There are open source MATLAB alternatives.

Great project. With effort it could be reverse engineered, but links to all the required files would make it far easier.

Seems the title of this post has changed since I first commented (it had referenced open source). Still, it would be nice to see all the design an operational files involved.

I’ve worked on $100k devices that are used for research. I believe there’s a good future for this type of work. Most of what makes those devices expensive is the R&D that goes into them, not the production costs.

That is somewhat true but there are some fixed costs that cannot be easily brought down.

1. Optics- especially when you start talking 63x (looking at internal cell structures) color corrected objectives are generally going to run around 10k, there is really no way around that. A microscope like the one shown here (and the GRIN lens miniscopes used in brains recently and the ball lens microscopes) don't require as good optics as they aren't imaging at super high resolution and color misalignment won't hurt them.

2. The camera chip/housing/frame grabber- Generally the actual chip is pretty low cost. But the system to cool and grab a lot of frames really fast is pretty expensive. The rates of commercial cameras-120 FPS at their highest, is completely insufficient for a research machine. Also the sensitivity of an 'off the shelf' camera is not going to be appropriate for a research machine. These costs hold these cameras around 5-10k. This project doesn't require these systems because they are looking at whole cells so they can afford to 'lose' a lot of photons.

3. Lasers- if a microscope has a laser (generally only 2 photons or certain types of confocals), it's going to cost a lot. I don't know what costs go into making a laser but the technology is pretty well established and the power of laser required for these machines seems to consistently run into the 20-50k range.

Other parts that are can be made cheap, but cost more (1-2k) when they need to be long lasting and highly precise. Stages (xyz stepper motors in particular), dichroic mirrors/filters, housing body, LED/fluorescent lighting source. This project got around that by specializing these components and using off the shelf stuff. I think probably many of these components could have their prices dropped a lot.

I think there is a future in bringing some of the costs down, but until we get a cheaper way to grab a lot of frames quickly or make good optics on the cheap, I don't know if research scopes (except the ones currently under patent like STORM) will drop much. Though alternative scopes (like the miniscope- 1k) can sometimes surprise people and replace expensive microscopes (120k two-photon), but that's more about being able to build into the niche and cutting out the need for more expensive components. Though from what I've read/seen microscope companies have a ton of overhead.

Can you elaborate on why optics are so expensive and why it isn't likely that manufacturing methods will come down in cost any time soon?

Only very high-end optics are expensive if your goal is to just get reasonable objectives you don't have to spend a lot of money it gets more complicated when you're doing things like laser illumination or if you need a really really low n.a.

I don't know the specifics for microscope lenses, I know a little about camera lenses, but I think it's largely high failure rate. It's a lot of mechanical grinding and heating and cooling with some final human checks. Each high quality lens has something like 10-15 lenses in it each which fails out pretty often. Also, shockingly, these lenses might still be hand assembled and checked? (I can't find a modern reference on this).

As to why it won't get cheaper anytime soon a couple reasons (at least what I'd guess):

1. Small batches and minimal incentives. It's feasible the high end microscope lens market just isn't worth disrupting. New processes could possibly make better lenses cheaper but just setting up the fairly massive tooling probably would eat up the margins. These lenses last forever so the yearly sale of these lenses is actually pretty small. (Funny enough there has been recent stories where competing manufacturers have started buying up some type of lens and actually completely drained ALL the available lenses of that type on the market, and that was a fairly cheap lens). Also people buying these lenses- labs and biotech tend to not care very much about price, so there is little incentive.

2. Replacement materials don't exist yet. The easiest way to beat the lens market would be to replace the glass lenses with plastic. This is what happened with phones/disposable cameras and is why camera prices plummeted. But at present this hasn't happened for higher end lenses.

3. There hasn't been cheap increased precision in machine movements really. So failure rates are unlikely to go down. If someone could adapt lithographic techniques to lens making it's possible it will get cheaper?

4. Kind of a dumb reason, but the biggest reason they aren't likely to get cheaper is that they haven't gotten cheaper. The process is about as automated as it can be and the microscope world is on the cutting edge of a lot of technologies (photolithogrpaphy, novel microscopy techniques, etc) if applying some technique like ML to some part of the process could make it cheaper, someone probably would have pushed it. There are also a lot of DIYers in this area. It's not a technological stagnant market. (I know this kind of competes with reason one, but what I'm saying is there is not 'easy win' to apply here.)

5. The big research in microscope/optics design isn't really focused on making high end lenses cheaper (As far as I've seen). It's more focused on stuff like making even higher end lenses that let you cheat the rules of optics, or like this paper, using low end lenses to do experiments that were impossible with more expensive lenses, making tiny/embedded systems, using beefy automated systems to image a ton of stuff simultaneously (can image a whole rat brain in like an hour now), and automating capture speed with the systems that exist (if your 100k system can image 10x faster and doesn't need a user it's like having a bunch of cheaper systems).

In summary, it's a well explored, small market, that no one is throwing money at and no obvious tech advances touch. Though this is largely speculation from talking to microscope reps and wandering around vendor fairs/automation conventions. Also I'm a biologist with an interest in optics not a microscope expert.

$250 in parts is not a $250 microscope.

The expertise required to design, source, align, calibrate, stock, distribute, sell, and support such a microscope is not without cost.

But usually nobody quibbles about the same sort of definition for say a $250 3d printer (which requires all that fiddling too)

The difference being that professionals don't generally buy $250 3d printers - and when they do, i expect they do factor those costs in.

The difference is that professionals get paid and they know the value of their time. If you cannot rely on free labour, factoring in associated costs is important.

They would if people were producing 3D printers for African villages

It seems like there are a mountain of health related challenges that would benefit from some application of a little tech. I came here to ask what were some other hacking opportunities, how to find them, and how to collaborate with the med/bio people that need it.

Billg funded this project, and of course he's got something to say: https://www.gatesnotes.com/Development/William-Wu-changed-hi...

You can find those challenges yourself by volunteering your time at the grass roots level. One of the top of my head - One of the challenges in mass immunization (especially in poorer countries) is that the 'last mile' data is very sketchy and mostly self-reported. Many times people just have to assume that immunization campaigns are successful without proof. It would be nice to create a device/method/protocol that could automatically collect data, like say who was immunized, what were they given, number of doses, etc, etc. Conceptually the challenge looks to be easy, but in practice at the ground level is where the real challenges lie.

Another idea is creating a smartphone app that end-users can upload test results via the camera. for e.g. in the case of tuberculosis, once you administer the test, you have to observe a color change on your skin within a certain period of time. The key thing here is that in places where the number of doctors are few, we want to create a system where the patient who is taking the TB test only goes to the doctor if they actually have TB. A smartphone app that can deal with various lighting conditions, various resolutions, various color casts (due to auto white balance issues), etc, etc would be an interesting option.

The paper supporting this work: https://www.biorxiv.org/content/10.1101/684423v1

It scans blood thin smears for malaria; but scans sputum samples for Tb.

Cheers Manu Prakash

Can you compare this approach to the origami and spinning projects for malaria detection? Are they complementary?

I was thinking about an all-in-one device like this, naively wondering if it would work to throw many phases and frequencies of light at the sample, and train AI on that.

Though this microscope has specific modules for tb or malaria, I wonder how well those filters would find other things with some ML.

If only this were out before the fall of Theranos, they could have stuck it in a fancy box with a robot and called it "Einstein".

I can easily say that this happened despite theranos.

Sounds like a promising approach. Maybe they should ask Gates for some funding.

They are funded through the gates foundation

https://twitter.com/PrakashLab/status/1144453793485955081

Wrong. It's a $250 microscope. Any other cost is culturally imposed. Values are not objects.

Time isn't free.

It is given to us freely everyday without asking. What we do with it determines its value.

If you create powerful microscopes with that time, what value would be determined?

I think everyone has 3 numbers. A number that they will not go below, a number that two parties can agree on for a transaction, and a number which they would be extremely happy with (with no upper limit obviously). No different than negotiating a salary / contract which is what it is. I think when someone designs something with the intent to sell it however, that cost of research and development gets baked into the expected returns on the volume of sales. Its like an artist who sells originals versus an artist who sells copies of artwork, if that makes sense.

Every hour costs us one hour of our lives.

microscopes (hand-built ones like this) require a great deal of additional manual labor on top of assembly. It's not culturally imposed- you need to master a ton of important and complicated optical stuff before this device works well.

Please don't take HN threads on generic ideological tangents. Those are less interesting than whatever they tangent off from. This is in the site guidelines: https://news.ycombinator.com/newsguidelines.html.

We detached this subthread from https://news.ycombinator.com/item?id=20305597 and marked it off-topic.

I have Twitter blocked. Is there another source?

From the twitter post:

https://www.biorxiv.org/content/10.1101/684423v1

Reply on news.ycombinator.com