This week I am talking to Micha Breakstone, (@michabreakstone) CEO and Co-FOunder of Neuralight.ai (@neuralight) a company using novel objective and sensitive neurological biomarkers to accelerate neurological trials and increase their success rates
Micah has an unusual background and history who thought he would be a novelist but found his way into serial entrepreneurship successfully launching Chorus.AI, a company focused on analyzing conversation unlocking the hidden dimension of insight found in every interaction we have with others.
Thanks to this background and meeting his co-founder he formed a company that is determined to bring science and objective measurement to the world of neurology. As he (and I) both agree
You cannot improve what you do not measure
As Micah points out with over 1 billion people worldwide suffering from neurological disorders like Parkinson’s and Alzheimer’s but those diagnoses and treatments fall in the shadow of subjective assessments that suffer from wide variability and even intra and inter-rater variability making real progress difficult to achieve and even harder to measure.
Listen in to hear how there is a better way, one that might allow for easier more accurate diagnosis of neurological diseases and offer tools to assess the efficacy of therapies accurately by digitizing neurology with oculometric measures from facial videos captured with a standard webcam or smartphone and using AI to accelerate clinical trials and drug development and improve patient outcomes.
Listen live at 4:00 AM, 12:00 Noon or 8:00 PM ET, Monday through Friday for the next week at HealthcareNOW Radio. After that, you can listen on demand (See podcast information below.) Join the conversation on Twitter at #TheIncrementalist.
Listen along on HealthcareNowRadio or on SoundCloud
Raw Transcript
Nick van Terheyden
And today I’m delighted to be joined by mica Breakstone. He is the CEO and co founder of neural light AI. Mica. Thanks for joining me today.
Micha Breakstone
Thank you so much for having me, Dr. Neck and you got the pronunciation of my name perfect. So I’m impressed as well.
Nick van Terheyden
Well, I appreciate that. I do try because people certainly mess up by me. Okay. I’m very sympathetic. So, as I do with all of my guests, I think it’s always important to get some context, you’ve had an interesting career. You’ve had some successful past exits. Tell us a little bit about your journey and how you got to this point, if you would,
Micha Breakstone
yeah, thanks for asking. So I think until the age of 29, or 30, I thought I would become a famous novelist. One day, I was in the middle of doing my, my PhD. Part of it was at MIT, and I saw this sign that said, if you’re not embarrassed to show us your demo, you’ve waited too long. Like, oh, that’s a cool way to be creative. And I launched my first startup, which was not a great success. But I got the bug within me and I formed company after company and reach some really nice successes. You know, formally, my background is a master’s in math, and then a PhD in cognitive sciences. So I’ve always been interested in how the brain works and how cognition works. Relationships between the brain and and the world, itself. And my previous company was a company, we ended up selling for around $600 million. We analyzed conversations, the goal there was to unlock the hidden dimensions that govern decision making and voice conversations, in order to understand and enhance human communication, at least that’s what I told myself, we ended up helping people sell better, which was also amazing and lucrative, but not really a world changer. So, you know, after chorus, we made some money, we had a nice secondary, I started angel investing. And then I was lucky enough to be introduced to my current co founder who told me about this huge unmet need in neurology, and I was like, Oh, wow, this is actually an opportunity to not only, you know, build an amazing company, but actually change the world for the better. So we dug in, and obviously, I didn’t have a lot of background at all, in, in neurology, or in healthcare at all. But the theme that we were going for was, you know, if in course it was, you can’t improve what you don’t measure, we measure the sales call here at near light is you can’t develop cures for drugs for diseases, you can’t measure and we measure neurological disorders. So it seemed like an amazing challenge. And we were able to group around us some amazing human beings and, and learn very quickly. And so that’s kind of where we are. And I really hope that your life becomes my magnum opus, our magnum opus, in that we are really able to transform the way neurology and precision medicine for neurology is done today.
Nick van Terheyden
Yeah, so it’s interesting, you bring up that quote, because I use that frequently in healthcare, especially because I think we’re challenged with that across a number of spectrum. It’s variously attributed to different people I attributed to Lord Kelvin, you know, I’m a little bit biased with a British accent, you know, the 1800s. I’ve had people argue that the fact that maybe it was prior to him really doesn’t matter. But you’re exactly right. If you can’t measure it, it is very difficult to improve something. And before we get into neuro Lytx, tell us a little bit about what you were measuring and what you were doing in this previous company.
Micha Breakstone
Yeah, so Well, of course, is again, it’s very far we were measuring conversations, right. So we were looking at everything from, you know, what were the most important semantic cues? What were the sentiment that were measured? Were there any unseen patterns that were that would give away? What was you know, happening in the conversation, we would look to automatically surface differences between great sales rep and mediocre sales reps and bad sales rep. So just give us an example. You know, going back when it was like, Do Is it true that a great salespeople ask more questions than less better salespeople? And you know, our intuition was, of course, they asked more questions because they want to get more information. And when we dug in and let unleashed, quote unquote, artificial intelligence or AI, we found that our intuition was wrong. All salespeople ask the same amount of questions. It’s not about the number of questions. It’s about the quality of questions and the type of questions so great salespeople, ask what’s called open ended questions or questions that elicit longer response times and let people actually uncover their pain He’s in talk openly. So those are the kinds of things that we did that car is another example being, you know, setting action items, we would have thought that terrific salespeople set more action items. But in fact, that’s not true. It’s about when you set the action items, terrific salespeople, set them way earlier up in the conversation so that when they end up at the end of the conversation, you’re just recapping something you have already agreed on. So those are kind of things that we surfaced kind of automatically, which really, you know, is a nice analogy to what we do at neuro Lytx. Namely, let the data speak for itself. The way I most like to bring this close to home for people’s thinking, because most people aren’t really acquainted with neurology, but almost everybody knows someone who suffers from some sort of, let’s call it depression. When you go into a psychiatrist or general practitioner and you’re supposed to be suffering from depression, they’ll ask you, you know, a battery of questions such as, you know, how’s your appetite? How’s your sleep? How’s your libido? Do you have any suicidal thoughts, etc, etc, etc. And after about 20 questions, they’ll say, Well, it seems that you have depression, or it’s an MDD, or it’s increased or not decreased, let’s change the dosage, increase it, reduce it, whatever. And it’s like this, right? It’s entirely Sorry, I was pointing with my finger, and I realize it’s on the audit. So it’s basically just a guestimation of what is happening. And in fact, the exact same situation happens in these other horrific diseases such as dementia, and multiple sclerosis, and Parkinson’s and even ALS. So being able to actually not use only this clinical observations, and not only these questions, which are inherently subjective, by definition, actually have something like a 25 to 30% interrater variability, meaning to doctors looking at the same patient, same day will disagree with each other 25 to 30% of the time, in fact, and even more ridiculous fact, there’s a 15% intro rate or variability, same physician looking at the same patient, same time, not unbeknownst to him, that it’s the same patient will disagree with themselves 15% of the time. So insanely subjective, extremely non sensitive measures. So no with neuro light, it really is about heralding in this these objective insensitive measurements or biomarkers, as we
Nick van Terheyden
call them. So I think, you know, the important point here is that, you know, that’s subjective analysis. And, you know, let’s just cream offer a decent insight, you know, incremental insight from from this, that, you know, the top three list of sales activities that will win you more business is complete junk, because it doesn’t drive into the detail. And I think you beautifully articulated that, in the, you know, people go ask more questions, you know, that’s probably top line item on guidance for new salespeople. And I love the sort of dive in so it sounds like, you know, that was as much about natural language processing. This wasn’t about voice so much. Were you analyzing the tone and the cadence and so forth? Or was this about the words?
Micha Breakstone
Of course, of course, we did both, actually. So we did both tone analysis sentiment as it pertains to voice and and the semantics themselves, there are various different phenomena that are really, really curious. So people, when they’re less confidence in different depending on the culture, they will either end with a higher tone or a lower tone, the more you have an executive presence. The question mark at the end, right, is it? Well, you’ll increase the tone a little less. It was also about, you know, how how fluent your voice was, how many times you had these kinds of Futter words like you know, and, and I mean, and stuff like that. So it’s also about the inflection of the voice, the use of how often you cut yourself off how fluid your voice was. So we analyzed everything, we ended up finding that there were very interesting cues that really separated the very top from the middle, and then from the middle, and the lower part. And what was really cool at the time is we automatically surface key points for the people in the middle of the batch to be able to improve themselves and surface these key moments that they could train on and maybe share with their managers to get to increase the impact maximally for them.
Nick van Terheyden
So I think great insights and you know, clearly from from all of that data, just fascinating. I mean, it feels like you could have had a whole career that just focused on improving sales based on the insights that you derived. Before we get into neural Lytx and you know what you’re doing there. I think it’s worth explaining, as best as you can Why the application of artificial intelligence in this instance is so important because as I listened to you, there’s a part of me that goes, well, I could have observed that. But I think the reality is that that’s not true. Can you explain that and give people some context?
Micha Breakstone
Yeah, for sure. So first, you know, AI means a lot of things to different people, right? There’s a meme going around amongst intrapreneurs. Mostly, that says something like, if it’s written in PowerPoint, it’s AI, if it’s written in Python, which is a coding language, it’s machine learning, right? So often when I hear AI kind of shutter because I want to say, Wait, we’re not charlatans, right, it’s not only about throwing in a lot of data and hoping that something beautiful comes out not at all. We were deep into the science and we try to understand also the mechanisms not only correlations, but also causations. So, you know, we do use AI and mostly also signal processing, which is a type of AI a little a little different than standard machine learning to uncover these patterns and be able to extract signals in a very precise way. Indeed, as you you know, kind of insinuated or hinted bedside, evaluation of eyes has been, you know, part of the clinical evaluation for, you know, since neurology was born, right, you know, all of these diseases as they progress and mass PSP, Parkinson’s, they all they all do present themselves in the eyes, right, you know, even looking at different Parkinsonism. So, if you look at the difference between PSP and Parkinson’s, or progressive Supranuclear, Palsy and Parkinson’s, you’ll see in PSP, you’ll see some kind of hyper metrics mechanic movement, namely, and people not being able to actually move all the way up the eyes horizontally, and also kind of on the vertical gaze, they won’t be able to gaze downwards. So you see this in standard neurology today, right? Somebody reports that they have Parkinsonian symptoms, but they fall down the stairs a lot. Well, that’s probably because they can’t look down and don’t see where their stairs are. And although PSP and Parkinson’s look very, very similar early on, you might suspect this is actually PSP and not try to, you know, prescribe some dopamine dopaminergic drugs, because it just won’t work. Right. So these clinical assessments have and will always be an important part, but the essence of using quote unquote, AI and I say, quote, unquote, because it’s important to say this isn’t a generic blanket term, but rather a deep understanding of the signals. That’s what I mean here by AI, both in uncovering them and in analyzing them is really important to uncover these things a with much better precision be much earlier on See, in a continuous way, in order to both diagnose and monitor the progression thereof and help with the treatment, the dosage administration of the drugs as you develop cures for these drugs, if that makes sense.
Nick van Terheyden
So for those of you just joining, I’m Dr. Nick the incrementalist today I’m talking to Meeker Breakstone, he is the CEO and co founder of neuro Lytx AI, we were just covering off the background to AI and the way that that process is applied, you did so successfully with sentiment, content analysis, intonation and so forth for voice in a previous instance. And now you set your sights on this healthcare space. And, you know, we sort of established that the foundation that to be able to improve things we have to measure and we don’t have good solid measures for many diseases. And in this particular instance, we’re talking neurological diseases that have very widespread impact. Help us understand what it is that you’re doing and how you’re going about that neuro light.
Micha Breakstone
Yeah, so at the very essence of things what we’re doing is we’re using standard video on namely taken captured by you know, the webcam, your audience won’t see them but as we’re using now the standard webcam or even a smartphone to take regular video and then use those videos to extract hyper accurate even sub pixel resolution features of these I’m movements and put together a signature of every neurological disorder and a signature that pertains also to the indication itself and also to its acuity, namely, how much it’s progressed along the clinical scale. So for for example, you know, if it’s if we’re looking at indication like Ms. So Ms. may look quite similar to other autoimmune CNS diseases, but we’re able to actually extract a specific signature for MS and then the standard gold standard that the FDA uses called EDSS. Usually, in normal clinical trials, it usually takes about two years to see a movement on the EDSS scale, were able to tell a statistically significant change in the progression of the disease within three to six months, so much earlier on, so much more sensitive, obviously repeatable because it’s, you know, its objective. And, and very much telling of the disease itself. So basically, you know, in a nutshell, we’re taking standard video extracting subpixel parameters, and creating the signatures a of the indication B of the acuity or the progression of the disease.
Nick van Terheyden
So it’s, as you analyze this, is this confined to ocular movements that the movements of the eye or is it broader than that? Are you looking at a whole face and other elements?
Micha Breakstone
Yeah, so for now, we’re focusing exclusively on ocular metrics. And the reason is, there’s a huge body of work, namely, over 750 papers published over the last 20 years, showing that these correlations between micro movements in the eye and the progression of a wide spectrum of neurological disorders is extremely robust. What we’re doing and this pertains to your previous question about AI, we’re not limiting ourselves to one paper, one set of parameters, rather, we’re looking in one fell swoop, we’re extracting over 100, different ocular parameters and putting together the signature and we believe that this will be more than sufficient to measure to both diagnose in a very accurate way, what the what the disease is, and create a very strong, actually a novel standard of measure for the acuity or the progression of the disease. So for now, we’re focused entirely on ocular metrics, although some of the studies we’re joining have, you know, different biomarkers, which are fine. We don’t, it doesn’t bother us, but we really believe in the power of ocular metrics.
Nick van Terheyden
So this is clearly exciting. I think there’s some tremendous potential to be able to, you know, at some point, even arrive at a diagnostic capability that would, you know, as we’ve seen, in other instances of camera use case, being able to apply intelligence and deliver diagnostic, where are you on the FDA process? Because they’re they they have some very strict regulations to cover these kinds of activities?
Micha Breakstone
Oh, for sure, look, the FDA is going to be a very long journey, right? It’s not, we’re not, you know, if we’re if the end goal is to become a surrogate endpoint for the clinical scales that are used now, that’s gonna take, you know, say five years and give or take right a year or two. We’re in the process of submitting our for initial set of data, which we’ve used for both healthy patients over 1000 healthy volunteers, and then a significant number of of patients suffering from indications. That would be you know, the standard processes, you submit a 510 K to show non inferiority based on you know, comparing to a specific measure that or device that is used today. And then, as we participate in more and more trials with these pharma companies that we’re collaborating with, we’ll be able to send in more and more of these clinical evidence that what we’re doing should be considered as a proxy, or as a surrogate endpoint. And that will take time. And for this reason, you know, we’re not going directly in selling this to clinicians, right, or the, you know, to patients, the go to market that we have in mind is partnering with companies that are developing drugs and the pharma companies in order to help design trials that will be much more successful, more efficient, more effective. I don’t know. I don’t know you’re probably aware of this. But CNS drug trials have less than a half of the chance of success than non CNS drug trials, which is incredulous. And often these fail just because of bad design. So to give you an example, Parkinson’s, early, early diagnosis of Parkinson’s is usually 25 to 30%. Wrong. So if you have a cohort of 1000 patients undergoing a Parkinson’s trial, you have 250 that don’t actually suffer from Parkinson’s. They’ll have you know, essential tremor or, you know, PSP or or some other, or MSA or some other Parkinsonian ism disorder that is early on very hard to tell the difference from Parkinson. So you end up you give this drug, let’s say it’s really a good drug. So you see a lot of respondents, but you also see 250 that are complete noise because they don’t have Parkinson’s to begin with, right? So in order to overcome those, that the effect has to be extremely strong, usually, it’s not as strong as you would like. So you end up not achieving statistical significance. So the question is, well, what do people do today? Well, they either throw out the agent the molecule, or have to redo the trial entirely. With neuro light, basically, you have this tagging of data saying, well near like, suspect, this person doesn’t have Parkinson’s, but rather say essential tremor. And then you can reanalyze the data, because you’ve defined it ahead of time without being considered cherry picking, or P hacking or whatnot, right? So you literally increase the chances of success, basically taking into account a sub optimal design of a trial. So those are kind of the things that we’re able to sell way ahead of any FDA approval, right. And, you know, submitting the 510 K is great, because it proves that at least we’re measuring what we’re saying we’re measuring, right, you know, and in fact, I can share that our early indications we measure ourselves against these these dedicated hardware devices, I won’t name the names of the companies, but you know, machines that cost $30,000 In our you have to put your head on the chin rest, and sometimes even kind of hold your head in one place restrain it, we’re actually we’ve shown that we’re within a 1% error margin of these machines. And not only that, but our the variability we have is much lower. So that probably tends to say, not only are we not inferior, we’re probably actually superior. And we’re not using hardware to do this. We’re using software, so namely signal processing, to do that.
Nick van Terheyden
So I think clearly exciting opportunities. Exciting times, I think there’s many sort of strands, as you think going forward. You know, I listen to this and think, Well, you’ve got ocular movements. But you know, I imagine throwing in some other data, potentially, as part of that, maybe you’re gonna start to see and create some insights that we don’t have at this point. Where do you see this going? And what are you excited about?
Micha Breakstone
I’m excited about the future in reinventing neurology, right, the way I see our company. You know, the previous company was sold for around 600 million. I hope this one IPO is for at least 10x. That I think neurology is an absolute huge domain, hundreds of billions of dollars are spent on on developing cures, that the market is insane that the world is getting older and older. And so these neurological disorders often are showing themselves much more the prevalence is more than doubled, for example, for Alzheimer over the last 10 or 20 years. So I see ourselves as introducing, are ushering in this really this this new standard of evaluation and of care and monitoring of Neurological Disorders and, and ushering in this new era of precision neurology, precision medicine for neurology, I really know this company will be a success, if 10 years from now, we will help develop and potentially even be part of these new drugs, as in a, you know, as a companion type diagnostics for these really targeted novel drugs that help solve and cure these terrible, terrible neurological disorders.
Nick van Terheyden
So I think exciting times plenty of opportunity, as you’ve said, almost, I don’t want to say Greenfield, that would be a little bit disingenuous, but certainly wide open in terms of opportunity for filling in the gaps over and above any kind of subjective assessment that we struggle with in, you know, this specific area. But I imagine in a number of other areas, potentially not sort of defocus what your current direction and intent is, unfortunately, as we do each and every week, we’ve run out of time. So it just remains for me to thank you for joining me on the show. Mica. Thanks for joining me on the show.
Micha Breakstone
Dr. Nick, thank you so much for having me. It was a pleasure. And thank you for the very thoughtful questions and I hope this helps your audience
