MA: I’m going to talk about Deliberate.AI and what we are doing to focus on the problem of subjectivity in mental and neurological health. So it’s one of those final frontiers of medicine where it comes to how we’re doing diagnosis, how we’re thinking about treatment. We’re really relying on subjective measures. We can’t currently look at an x- ray to give you a definitive view of what’s going on with a patient and what would be the best course of treatment for them. And that problem of subjectivity crosses diagnostics, treatment, and clinical trials. So in diagnostics, we see the problem of actually many diseases are only picked up later in their development path. There’s a high degree of misdiagnosis, about 64% of common mental health problems are misdiagnosed in primary care. And it’s an inefficient process. In treatment, we’re seeing the problem of trial and error for mental health specifically. We just start some course of treatment, hope it will work, and if it doesn’t work, we’ll switch it out. There’s a lack of monitoring. And if there is monitoring, it’s often based on self-report skills that only 1 in 3 clinicians uses, and only 1 in three 3 fills out. So overall, we see a high degree of limited remission, really, and a lot of these conditions become chronic. And in clinical trials, we have the problem of very low FDA approval rates for new drugs. And to make that even worse, CNS trials are twice as likely to fail in Phase 3, so the end of the cycle. And what we also see is a large heterogeneity in response to treatments. And we don’t really have good ways of enriching trials to focus trials on those that are likely responders. So what are we doing with Deliberate.AI? So we’re in a golden age of AI, but I think what we’re recognizing is that we really need multimodal AI methods to solve this problem for mental health. So we’re using multimodal AI to get more accurate diagnostics and to get more decision support information for clinicians and clinical trials. Now, how do we do that? We focused initially on the key thing that’s used on specific to CNS indication, which are behavior. So we take video signals, and from that, we extract the most comprehensive set of features. So we look at people’s facial expression, we look at their head movement, we look at their eye movement, we look at kind of the way their voice sounds, like the way their muscle are impacting the way their voice sounds. We look at kind of the words they’re using, the way they form their sentences, as well as things that we can pick up, peripheral physiology such as heart rate from video. Now, there are several companies out there that look at these things individually, but what we do a little bit differently, we’re actually combining them. So there’s a lot of scientific evidence linking these individual biomarkers to a host of mental and neurological conditions, but their predictive power is often limited. So we’re actually combining these in multimodal models that actually allow us to do three things. So the three types of models we create are around diagnosis and monitoring, which we call AI-COAs or AI-derived clinical outcome assessments. Predictive decision support, and we’re now working on our first model called SHERPA, which is Suicidal Hazard Evaluation, A Risk Prediction Algorithm. And the third one is actually around clinical process support. So it’s not only what are the symptoms of the person, what treatment are they getting, but also how is that treatment delivered to them? And we’re there kind of now automating ways of assessing therapeutic alliance. Now our pipeline really focuses initially on these kind of diagnostic and severity models, and we’ve kind of now in pivotal trials to get depression and anxiety models regulated for both drug development tools with CDER, as well as de novo classifications of medical device to be used in clinical care. We have some other indications kind of following soon. Now how do we get those audio-visual signals? So our business model is primarily focused on APIs. So if you’re an app developer or if you’re a healthcare provider, come talk to us. We’re looking to kind of make these APIs available for you to integrate in your workflow. But we also recognized there are ways for us to kind of serve those that don’t have those complex or sophisticated workflows yet. So we have developed a platform that allows us to kind of observe mental health interactions, and from that, do a couple of things. We help reduce workload from automatic transcription and using AI to kind of summarize notes, but then also use these models that we’ve developed to give you a sense of symptoms severity. So rather than having to rely on patient self-report, we can track over time how patients are responding to treatment, and even indicate and identify ways in which you want to intervene perhaps for participants or patients that are not on an improvement path. That product has been piloted over the last six months with the largest healthcare system in New York and about a dozen or so community mental health clinics over the last year. We’re now gearing up for public launch. So if you’re interested in participating in that, if you’re a provider, feel free to kind of use this QR code or kind of link with me after to give you access to that. The other way is actually kind of true interaction and taking the whole interaction away from the need for a clinical interaction. So we’ve completed a beta test using kind of an avatar-based interaction that allows us to kind of, rather than kind of patients filling out surveys or kind of have a lengthy interaction with a clinician, how do we take some of that burden away from a clinician and do that before they even get a meeting, and help healthcare providers kind of determine what’s the best triage for this particular patient based on that interaction. Now, what are the benefits of all of this approach? So kind of really what we’re focusing on is both kind of how do we improve healthcare with in the sense of more efficient and accurate diagnostics. How do we kind of support the move to value-based care? How do we move away from subjective self-report that many kind of patients aren’t filling out? And to kind of help with this transition of both understanding how patients are improving as well as like kind of giving information to clinicians about ways they can improve their treatment to ensure outcomes are improving. And lastly, certainly not least, that we’ve made most of our progress currently in how do we kind of enable powerful and more faster clinical trials. So that starts with efficient recruitment rather than having to screen 200 patients of which only one kind of meets your enrollment criteria. How do we do some of that in avatar-based prescreening? The most important thing is around sample size reduction. So using our AI-COAs for depression and anxiety, we can reduce clinical trial size for depression and anxiety trials between 10% to 30%. And we’re kind of on track to be the first AI tool to be in the novel pathway for the FDA for innovative science and technology for new drug development as a drug development tool. And the ultimate goal, and we’re starting to have some kind of collaboration on that, how do we kind of focus on enrichment? How do we increase the effect size by enrolling the best patients that are likely to respond? Now, as was mentioned by Chris from Qlik Therapeutics yesterday and many others, like everything for digital health companies currently is about how do you actually create, generate the evidence? So we’re very kind of happy to have an extensive collaboration with several medical and other institutions to kind of really drive that evidence base, both to develop models and then independently in prospective confirmatory trials, demonstrate that they work in a real world context. We’ve got some great awards, including from HITLAB here that’s been helpful in moving us forward, as well as some accelerator and startup memberships. A call to action. If you’re interested about what we’re doing, feel free to kind of follow us on LinkedIn. If you’re a provider interested to think about like, hey, how could this technology help me? Feel free to kind of click up on our link and get on the wait list for a public launch or reach out to me. And if you’re developing new products where you think this technology could be helpful, also kind of reach out. Finally, we’re starting our next fundraise next month. So if you’re a qualified investor, feel free to speak to me and happy to tell you more about that. That was my presentation. Happy to take some questions if helpful.

JA: Round of applause for Marc. We will take any questions from the live audience if anyone has any questions for Marc and Deliberate.AI.

MA: I guess I told a comprehensive story.

JA: If we don’t have any volunteers from the crowd, we do have some questions from Zoom, Marc. So Victoria from Zoom asks a data security question. So how have you implemented patient confidentiality into the platform?

MA: So there are a couple of ways in which you’re doing that. Obviously, all our platform is HIPAA, GDPR, and FDA Part 11 compliant. That’s more about ensuring the infrastructure is secure. There are two things that we can do. We’ve developed an algorithm, a GAN algorithm, that actually can de-identify audio-video recordings of patients and retain the features that are relevant for our models. That’s one thing. And we’re actually working on, which will be another nine months, to actually do what we call real-time feature extraction, which means that we can observe an interaction without the need for recording. So there’s no ongoing recording of a patient. The features are extracted in real time. And by doing that, we can inform our models and give us the outcomes. So kind of ultimately, we’re getting to a point where there’s real privacy enhancement and there’s no need for ongoing biometric collection.

JA: Anybody have any questions from the crowd? We have time for one more, if anyone in the crowd has a question for Marc. If not, there is another from Zoom for you, Marc, as well. I think Dan here says, what was the inspiration behind creating the software that aims to diagnose patients with the psychological disorders using the highly objective methods you refer to?

MA: So we’re a team of really neuroscientists, psychiatrists, clinical psychologists, computer scientists, with the aim of solving this problem. So people have either had their own lived experience, have been exposed to the subjectivity in mental health care, to professional endeavors. It is just so clear that while there’s been a lot of money being invested in mental health care, if we really want to get to scalable solutions that actually impact outcomes, it’s not a matter of like, oh, I can speak to a psychiatrist more quickly. It’s about like, how do we do correct diagnosis? How do we really know whether a patient is responding? So all of that kind of drove us to say, well, rather than starting a company that tries to do something, let’s try to solve for this problem first. Because how do we know if a new treatment is working if we can’t accurately assess or reliably assess what’s going on with a person? So that’s really been our motivating driver. Yeah.

JA: All right. Wonderful. And Dan, I hope that answers your question. Okay, well one more time everybody if we give a warm round of applause virtually and in person to Marc.