@Scoonie I finally had some time to watch the EIQ meeting, and was left fascinated but a little underwhelmed as perhaps you were.

I still have no intelligence about the actual product itself - haven't been able to see it on an actual workstation, and see how it presents its findings to the reader. But from the interview I can gather a few points and questions

• EIQ relies on human interpretation of echo studies
• As @Strawman pointed out, the product uses the measurements as performed by the sonographer. In producing an echo study, the sonographer will come up with a bunch of measurements, which are spat out in a report by the software. It seems these are the data points that EIQ has built its model upon, and will in turn use to spit out its own analysis of an individual study
• In practice, the sonographer doesn't actually "perform" any calculations or measurements - the machine generates pictures, and they basically do low-fidelity colouring-in to tell the software what is / is not a structure of interest or a line-of-best fit around a curve
• Ultrasound/echo is operator-dependent, and open to a couple of major sources of error:
• Poor Colouring-in: will under or over-state the structure in question (I've sat in on meetings with sonographers (who all have actual years of training) debating which shadow is or is-not relevant when drawing their lines
• Poor image acquisition: due to time-pressure, fatigue, skill, training, or patient difficulty, its possible that the image won't be collected properly, and measurements will be wrong
• Many important measurements rely on analysing speed of blood by using the doppler effect. Using the doppler principle, if one gets the probe in perfect alignment with the direction of blood flow, you can measure speed perfectly. But if off by only a few degrees in either direction, the measured speed will be less (and the lesion understated).
• This isn't a bell curve of accuracy, where one may hover above/below a mean - measurements can either be perfect, or will give a lower speed and diagnose less-severe disease
• Echosolve SHOULD only Look at Measurements
• Relatively easy to incorporate large volumes of numerical data into an algorith
• Relatively difficult to teach a machine how to correctly differentiate between different types of shadows on an echo image (takes humans years of dedicated practice to learn).
• Outsourcing the image processing to a trained human is a logical and reasonable step and much easier than teaching the machine how to do it
• Image Analysis by Machine WOULD be awesome though
• If used in real time, the machine could tell the sonographers that they probably haven't got the perfect image yet (I've seen this used in practice - very useful)
• If the machine is taught how to colour-in really well, it can probably do it better than a trained human, and generate better data / measurements (this was inferred very briefly, when he suggested that image analysis would be complementary to measurement analysis)
• In my experience, technology that aided this step would truly be value accretive to the trained sonographer, and achieve the best of human+AI combined that @Strawman alluded to in the early part of the meeting

• Echosolv Revenue is Potentially Vulnerable to Competing Solutions
• If Insurer CMS codes are based upon the insurer wanting providers to use the tech (because it saves money in the long-run), then EIQ is vulnerable to being outdone.
• If a better solution (in whatever form) comes along, will the CMS codes be cancelled / rescinded, or will it be up to individual institutions to decide which product they should go with?
• To some extent, all biotech suffers from this risk - a given treatment is only special if it delivers the best results. But in many cases, a cost/efficacy curve exists... the best treatment may attract a premium, while other treatments which aren't quite as good still add value
• If echosolv reimbursement is based on insurers WANTING institutions to use it, there is nothing to stop them preferring the Next-Big-Thing (should it come along)
• Do any Strawpeeps have insight into this process? Does the awarding of a CMS code provide a minimum/maximum period of reimbursement? If a superior product comes along, might some institutions continue to use Echosolv and its CMS even if it isn't reimbursed quite so well?

• So What?
• I do see the potential for this tech, but also see the potential for it to be superseded by real-time imaging solutions
• The IP is most inherently in the algorithm (which could be enhanced) rather than the dataset (which could be replicated)
• I can foresee uptake by the hospital system esp for workflow management, but don't see a sustainable competitive advantage
• A competitive and first-mover advantage does exist in the form of difficult-to-train algorithms on a large data set, but with multiple firms in this space and an explosion of AI computing, I am not convinced EIQ have a sustainable advantage here (although they may well convert first-mover into ongoing dominance if they continue R&D spend out of new revenue growth)

Example - Green line below is a human's drawing of a line-of-best fit around a doppler signal curve on an echo software. I bet a machine could do better at differentiating between important and non-important shadows

I use echo myself a bit in clinical practice, and I still dont understand what exactly the product does.

My best guess is that it is an app or a plugin to existing echo software (they all save echo pictures in the same format), which applies the algorithm to the echo loops (videos) to spit out a diagnosis or classification of lesion severity. My impression is (despite the company hype) this will be an aid to workflow / speed rather than diagnostic accuracy

Good points @Scoonie, and thanks for putting through some great questions for the meeting.

There really is a lot of potential here. I have no direct experience with the industry (obviously), but the addressable market and value prop seem very compelling. But then again, doesnt it always!

The hard point is knowing when a company like this goes from interesting tech to commercial success. It's genuinely made some good progress in the time since we last spoke -- but as Philip said, it's slow going by nature. I suspect it will be at least ~18 months before we get any early evidence of sales traction, and another 6-9 more from there to see if there's any momentum.

I put it into the same camp as EMVision actually. (And i have a tiny 0.8% holding there). It's one of those situations where it could easily 10-20x from here...or it never gets traction and it slowly erodes 90% of it's value over the next 5-10 years.

I think the way to approach these situations is to average up as the thesis plays out, with a preparedness to dump the lot if ever it looks like the vision isnt going to be realized.

To your question @nessy (on other straw), I took it that the next FDA approval will likely follow a similar timeline. So maybe a year at the earliest? depending on the data they can submit.