Slides from my AI Presentation at BIOMEDevice 2017 and introducing a new site

If you have been following the blog, you have heard I was presenting at the San Jose BIOMEDevice Conference this week. In connection with this, I have a new site, http://medpund.it/, focused on discussing my research in Artificial Intelligence, as well as to discuss the impact of the field on Medicine and Medical Devices. There will be discussions on technology, business and social aspects as well.

From time to time, I will highlight my posts from there on this blog, but I invite you to support me on that site, as you do here as well, by subscribing to updates using one of the means provided. I am trying to keep my focus areas separated on different sites. I would like to see how this experiment pans out. As always, you will not be bombarded with any advertisements, ever (maybe a couple of plugs here and there, but that's it).

Meanwhile, do head over to the site and download my slides, and provide me some feedback. The slides themselves, prepared in my style of minimal wordiness, do not tell the full story, but I will write posts on the various topics discussed in my presentation. Feedback is welcome.

The link: http://medpund.it/index.php/2017/12/07/slides-how-artificial-intelligence-is-changing-medical-devices/

Subscribe and Support, Please!

Did you enjoy this post? Please subscribe for more updates, using the sidebar. Have ideas or blog posts you'd like to see here? Contact me at yamanoor at gmail dot com.

Interview: A preview of my talk on AI and Medical Devices this December

I know I have not been blogging as frequently as I would like. It has been a busy few weeks filled with Maker Faires, conferences, competition judging and participation in competitions and panels myself. Just yesterday, I was a panelist at the MD&M Minneapolis Show, talking about issues surrounding cross-pollination of ideas in Medtech Development. Time permitting, I will summarize a few things I myself learned while sharing my thoughts and views. I hope to be able to blog again with regularity soon, hopefully, after my AI presentation.

On December 6, 2017, I will be presenting a talk on how AI is and will continue to affect medical device development and use. I gave a preview interview to MD&DI, Qmed on AI and Robotics and how they affect medical device development. The broad strokes interview, linked below doesn't address my entire talk, so if you are in San Jose, please tune in. In the meanwhile, do read the interview and let me know your thoughts.

Also, if there are any burning questions you would like me to answer, please leave a comment, or write me at yamanoor at gmail dot com . I look forward to hearing your thoughts!

The Interview:

https://www.mddionline.com/qa-how-prepare-ai-driven-future?ADTRK=UBM&elq_mid=1929&elq_cid=74447

My Presentation Schedule:

http://schedulebmsj.mddionline.com/session/how-artificial-intelligence-is-changing-medical-devices

Subscribe and Support, Please!

Did you enjoy this post? Please subscribe for more updates, using the sidebar. Have ideas or blog posts you'd like to see here? Contact me at yamanoor at gmail dot com.

Image Courtesy, Pixabay: https://pixabay.com/en/hand-robot-machine-697264/#

MDDI: Machine Learning in Healthcare getting some well deserved love

MDDI, a versatile source for healthcare industry news has some good news for AI/ML fans. This morning they provided news on two different companies that got funding for Machine Learning (ML) news. I cringe a bit when people keep confusing ML for AI. However, since ML is a focus area of AI, it is all good. I have provided the link to the article below.

Low Hanging Fruits

Healthcare cost prediction is tough. However, it is ripe for innovation with huge amounts of data and of course Machine Learning, now gathering steam through new technologies and paradigms. MDDI is reporting that Cardinal Analytx Solutions, here at home in Palo Alto has raised $6.1mn to do exactly this. I haven't had time to research the specifics of this company yet, having woken up to this news, but if I find something interesting, I will get back to you.

Analytics 4 Life, another company that is in the same article, is aimed at diagnostics, another area ripe for disruption through Machine Learning. It is apparently aimed at being a low-impact cardiac diagnostic, eliminating the need for radiation, angiography or other invasive techniques. This is intriguing, given how it will lower costs and be less intrusive to the potential patients. Accuracy is a concern, but reviewing some of the other ML based diagnostic projects through the O'Reilly AI Conference archives from NY earlier this year, I can say there is every possibility that a diagnostic can be created to be accurate enough using ML. Plus, when you are sitting atop $25mn from several investors, I say, you can do it.

Some Thoughts

1. I believe we are still in the pre-bubble stage of AI based investments in healthcare. This is really good!

2. Commonsense is still the norm in investments and that is also very good news.

3. Investments and start-ups are still mostly scratching the surface. It would be nice, if we saw investments in companies with deeper ambitions and technologies before the mad rush begins.

4. If start-ups don't pick up the pace, there will be an unfortunate stranglehold on AI by companies such as Google, Apple, IBM etc., especially on the analytics side of things. Overall, this is not good for competition or for innovation. Moreover, when the larger economy slows down, or at a minimum, when "tech" slows down, there will be a slowdown in the implementation of AI in Healthcare. This would not be good, should it come to pass, as, given political intractability and general cost increases are making healthcare expenditure rates go up.

5. Still no sight of innovation in medical devices or pharma, and this is also not very comforting.

There is of course, much more to be seen as data-intensive companies, use Machine Learning (ML), Deep Learning (DL) and eventually Artificial Intelligence (AI) and try to democratize healthcare. Of course, there will be a bubble, small or large to go through along the way, as it is in every changing industry, and yet, these are amazingly exciting times ahead! We are still taking baby steps!!

Subscribe and Support, Please!

Did you enjoy this post? Please subscribe for more updates, using the sidebar. Have ideas or blog posts you'd like to see here? Contact me at yamanoor at gmail dot com.

References:

1. The MDDI+QMED Brief: https://www.mddionline.com/two-ai-startups-bag-new-funds?ADTRK=UBM&elq_mid=1221&elq_cid=74447

2. Image, Courtesy Pexels: https://www.pexels.com/photo/man-with-steel-artificial-arm-sitting-in-front-of-white-table-39349/

Digital Health Companies have to be careful, otherwise they could spoil the game for everyone

NOTE: It has been a busy few weeks and it is going to be a busy few months ahead. So, my blogging frequency is going to be erratic at best. It is a shame, because I wanted to write on the Prostate Awareness Month and then on Breast Cancer Awareness in October. However, Maker Faire Exhibits, Conference Presentations, Panels and Judging lie ahead for me, so please excuse the varied publishing schedule.

SyncThink

MedCity Newes reported on a warning letter from the FDA to a burgeoning company, SyncThink and they seemed to have responded promptly, as one would hope. However, the tailwinds still leave behind a trail of caution for companies going forward. Here's a quote from the MedCity Article:

SyncThink’s device Eye-Sync was cleared as a prescription device for the purpose of recording, viewing, and analyzing eye movements to help identify visual tracking impairment in human subjects, according to the letter.

However, it appears their overzealous marketing folks decided to go ahead and bump up the device to include concussion detection as a "feature". Why not? What with studies about the NFL, and various sports agencies deliberately ignoring issues have rendered many sportspeople with permanent ailments. Concussions also happen due to accidents, falls, etc., and thus the available market obviously spans a wide range of ages and gender is no bar. And yes, they can be life threatening. All this, of course, does not mean that you need to put the proverbial cart in front of the horse.

SyncThink of course claims it removed the materials that offended the FDA immediately from all of its marketing mechanisms.

Good, Right? Well...

A Trail of Bad and Unacceptable Behavior

The history of Digital Health is rife with examples of bad behavior.

Let's start with Not-AI/Maybe-AI/Most-Likely-ML

Actually this behavior spans all aspects of Healthcare, including, IBM Watson, according to a recent STAT article - promoted as a cancer therapy recommendation system, and sold prematurely resulting in a number of problems. This is something I have suspected for a while now, and it has been confirmed. Jumping the gun, declaring poor, dysfunctional Machine Learning products, worse than what a teenager can do with TensorFlow has become the order of the day for some companies. And to see IBM, glorified by Satell in one of the iconic books on Innovation do this is just sad. It should make you wonder about others.

Real AI is really not here. Excuse the pun. Google and IBM have friendly monsters that do a great job at Deep Learning. The actual independent decision making that would be the functional result of actual intelligence (not algorithmic analysis output), simply does not exist yet, at least in the public eye. And most of what you hear, is really just Machine Learning. More on this in a few blog posts before my December talk at the San Jose BIOMEDevice talk on AI in Medical Devices.

The problem with packaging stuff as AI?

It will always be the same. Overpromise, under-deliver, dilute your brand AND make it worse for the people that follow you. Just read my blog post on renal denervation, when I complete it and get it out later this week. The World Faires saw fewer contortionists in all of their existence, compared to people still, desperately trying to peddle renal denervation.

And while Scott Gottlieb wants to make the FDA business friendly and what not, two things to keep in mind - the FDA was created to avoid the kind of mortality and morbidity seen in Europe because all kinds of snake-oil merchants walked in and did whatever they wanted, and second, Europe, after about 50 years or so, has finally decided to take a more FDA-like approach.

So lying, fluffing or overstating, will get you trapped somewhere with regulatory problems. And, then, there will be the face-saving that all the other regulatory agencies will do by trying to get in on the game: bashing you.

My recommendation: Just don't do it.

But, let's go through one more example, before we get back to the recommendation that you have, hopefully heard since you were a child.

23-and-you-maybe, but not me!

Years ago, when 23andme came out, I was very irritated and wrote unequivocally about how they were going to make things hard for everyone. Why? All they could do was process your DNA. However, they marketed it in a way to make gullible people believe that as long as you joined the mania, and got your DNA mapped, suddenly a cure for just about anything to be found. Again marketing, with a serious disregard of facts. I am sure at least initially, they were not able to meet their price point profitably (this is just an assumption I make), and were hoping to ride the hype wave of people thinking suddenly, all they needed was Microsoft Excel (or Google Drive, given the company's origins) and compare their DNA with a bunch of friends - and lo!, the answer to everyone's problems would be laid bare with genetic certainty. Well...

Back then (and even now, and for at least a handful of years to come), even confirming a Single Nucleotide Polymorphism (SNP), where a single pair of nucleotides is off its hinges on the DNA ladder, took months to years to confirm. So, the company was essentially planning to capitalize on the desperation of the sick. Well, long story short, it took YEARS for the FDA to tell the company to go pound sand (too bad, neither the FDA, nor I are allowed to use more colorful language).

The company shockingly enough first tried to push back, and eventually gave in. Remarkable. And along the way, they slowed things down for everyone else, who could have behaved themselves, and gradually introduced a usable product, legally, when its time actually came.

My Recommendations:

1. The Valley (there really is only one here, contextually) is full of "people", who think the rule of law exists for others. They quaintly call such behavior: DISRUPTION. Ask Uber how that is working out. You might tell me, well, AirBnB broke the rules and bullied everyone, and got away with it. I ask you, what is likely: will you be the Uber of Concussion Detection or the AirBnB of it?

2. Regulations exist for a reason. No, not that you may flout them and find your own Lucy Koh (of the Apple/Samsung fame), sitting on the Federal Bench, just waiting to bite off more than he or she can chew! Again, your business plan can't be, "let's burn the bridge and when we need to cross one, get acquired".

3. Don't ruin it for others. When you flout regulations, or put out a terrible, immature product (read about what Medtronic did with Renal Denervation), you screw over quite a number of companies. People lose jobs, patients don't get the care. Do you really want to be the geniuses that got everyone in that swamp?

4. Following the rules is actually not that hard. Hire people with experience doing this. Go to RAPS, or the ASQ and look up good people. And I got good news for you: "Talented people in medical devices actually get paid terribly". Compared to the so-so CS grad you hired, only to have to give him bags of money, feed him and do his laundry, so he may write misogynist memos ruining your already flaky reputation because you funded 23andme!

5. Regulatory Agencies are almost always behind the curve. If you doubt me, go ask a Software Quality Engineer, what it takes to provide a software-update on an FDA approved product. But they learn. In the meanwhile, do not try to hoodwink them because they don't understand the gibber jabber of AI and so on and so forth. These are really smart people, and prone to anger (if you had to read and interpret text only pdfs all your life - and NO, your ML algorithm cannot do a better job, you'd be on the edge too). Play nice. Explain yourself to them. Show them your intentions are good. And work with them.

Doing all of the things I have stated are not going to earn you brownie points. You are expected to do them. There is a lot of change ahead with Digital Health, 3D Printing, 4D Printing (yes, it exists, mostly in labs), ML/DL/AI, IIoT and who knows what. It is important that when applying these technologies and paradigms to healthcare, good judgement, patience and care is essential.

Subscribe and Support, Please!

Did you enjoy this post? Please subscribe for more updates, using the sidebar. Have ideas or blog posts you'd like to see here? Contact me at yamanoor at gmail dot com.

Reference:

1. On SyncThink: http://medcitynews.com/2017/09/fda-slaps-syncthink-warning-letter-concussion-detection-claims/?_hsenc=p2ANqtz-8nrGCq5sErFwlSoErAZCVWLTeQMaeWZlOm465jisv-uu2z4yqUt-sfcRu-YPPwD_LNPpWqAHnDZ-vxrsb-YPv7d0K6xA&_hsmi=56142139


2. IBM's Watson on STAT: https://www.statnews.com/2017/09/05/watson-ibm-cancer/?utm_source=STAT+Newsletters&utm_campaign=1793b79d98-Weekend_Reads&utm_medium=email&utm_term=0_8cab1d7961-1793b79d98-150030301

3. Image, Courtesy Pexels: https://www.pexels.com/photo/black-and-blue-electronic-tools-on-green-circuit-board-39290/

Now Medicine is plagued with NOT-AI companies, being called AI Companies!

There is a disease slowly spreading across many industries. It is not AI (and no, this is not a Luddite post, I am a huge fan of AI). It is NOT-AI, or Fake-AI or Pretend-AI - we don't have a term for it yet, and sooner or later, one will catch on. But first, let's take a look at the semantics and the problem. It is generally accepted that there are three levels of AI - Machine Learning (ML), Deep Learning (DL) and Artificial Intelligence (AI), progressively, more complex and capable. Variants exist, and yes, semantics are really important.

Machine Learning is a reincarnation of Data Mining, where algorithms help automate learning and response to new data without explicit programming. This is most of what we have as of now. Most "AI" tools, apps and companies you hear of, are limited to Machine Learning. Deep Learning applies artificial neural networks to go beyond Machine Learning algorithms to understand and analyze data. Deep Learning, will hopefully, eventually lead to a true AI that can use its innate intelligence to respond to data by analyzing it and make decisions based on it. This has not happened yet, except in some very narrow situations.

Like I said before, so far, what we have are algorithms, that look at data and perform meaningful, analyses and help us gain insights from them. Some might even make rudimentary decisions, but that is where things stop. Yes, Google's DeepMind and IBM's Watson are quite capable at AI, but they themselves are very good examples as to why, a lot of others claiming to be "AI" are not!

Here is an example:

Perusing through press releases a couple of days old, I came across "Mindstrong", a company that just raised $14mn in Series A Funding. I read about the company, and it is fascinating, but then I have to wonder who wrote the release and why they decided to distract from the clever things the company plans to do by mislabeling the technology/methodology as AI. I mean, you have the $14mn already, so why embellish it now? Moreover, the company is painting a very rosy picture of its clinical trial results, and while one hopes they are truly so, most of us, outside the investors haven't see any of these supposed results. So, this really is a stretch in multiple dimensions.

The company wants to passively collect data on the method of mobile phone usage employed by people with mental health and neurodegenerative issues and then improve treatments based on lessons gleaned from the data. This is great, a first step in using Machine Learning to improve patient outcomes, but without actual decision making, that would be at the same level as, or exceed healthcare decisions made by humans, it would not be an AI. This is not trivial. We live in a time where the FDA is yet to allow Clinical Decision Support (CDS) software to be less regulated. An approved AI is years away, at least.

Moreover, it creates unfortunate impressions in the minds of the public as to how far technological progress has actually been made in healthcare. While I understand start-ups ought to generate enthusiasm in the public and in existing and future investors, I don't think fairy tales will cut it. Moreover, the existing technology Mindstrong describes, which I am pretty sure is clever use of ML, is sufficiently exciting enough, it puzzles me that they would decide to engage in wordplay. An oversight in the press release, perhaps? That would not be so bad. If it is deliberate, then one has to start wondering about other things, such as the stated superiority of the clinical trials.

In general, I now worry, a trend I thought is a few months/years away is here now. Healthcare companies bandying about the word "AI" for things that are far from it. We need brutal honesty (the kind that lost Berkeley access to the CRISPR patents, but nevertheless, a very lofty approach - more on this another time) in helping us understand diseases, standard of care, technologies and all the limitations therein. Only through this, can we engage in disruptive as well as continual improvement of healthcare. Not, by embellishing press releases.

References:

1. The Release about Mindstrong Health: http://www.marketwired.com/press-release/mindstrong-health-raises-14-million-in-series-a-funding-2222154.htm

2. Image, Courtesy Pexels: https://www.pexels.com/photo/black-and-white-blank-challenge-connect-262488/