Israeli start-up cracks the language of the brain | Innov'Nation

[00:00:00]Now, everyday doctors can measure your heart, your lungs, even your blood chemistry. But, when it comes to the brain, many conditions like depression, PTSD, or Alzheimer's are still diagnosed largely through observation and questionnaires. An Israeli startup says it's changing this after 6 years in stealth mode. Hemispheric has unveiled an AI model that claims it can decode brain activity through a simple 15-minute non-invasive scan. So, for more, I'm now joined by the co-founders, CEO Hagai Lala Zar and CTO Gidi Litvin.

[00:00:36]Thank you very much to you both for being with me today. Um so, Hagai, let's start with you. We're starting with the pitch always. So, can you please pitch us what your startup is about?

[00:00:48]>> Thanks, Lynn. So, Hemispheric is an AI startup. We've spent the last 6 years developing, as you mentioned, a breakthrough model that's a frontier foundational model called Descartes for decoding non-invasive brain activity.

[00:01:01]And the problem is that the brain is the final organ in the body that we still only use subjective measures, and we don't have objective quantifiable measures to measure its function uh and its health.

[00:01:13]And, you know, if you think about AI's breakthroughs so far, things like LLMs, they translate the text of human language. The brain speaks an electric language, and our model, Descartes, what it does is it translates the electrical language of the brain into actionable understanding, into insights that can be used by doctors, by patients, by us to understand our brain function and our brain health and to make decisions.

[00:01:36]>> So, what exactly are you measuring?

[00:01:39]>> So, we use EEG is the sensor that we use. It's a non-invasive sensor. It's very mature. Um and it just measures uh electricity from the brain, from the scalp. And that is not the innovation of the company. We're not a hardware company. This raw signal uh is fed into from and our model being trained on huge amounts of of data, now understands brain function and can be used to build a suite of products in order for us to have quantitative information to make decisions, as you mentioned, on diseases such as PTSD, depression, anxiety, schizophrenia. These are uh psychiatric conditions as well as neurological conditions such as Parkinson's and Alzheimer's disease.

[00:02:18]>> And Gidi, what is then the edge that you have over doctors and normal scans, for instance?

[00:02:25]>> Okay. So, for most of the conditions uh we just discussed, there are no objective quantifiable measures, okay?

[00:02:30]So, mostly uh it's done via subjective uh methods, so questionnaires and and speech, right? So, what we do is we uh decode brain activity using our models to produce uh quantifiable measures.

[00:02:46]And our biggest edge is the data at the end of the day. So, we the build this model, the Descartes, which is a 6 billion parameter model.

[00:02:54]Uh we collect the data from uh uh lab chain, a network of labs that we've uh established in Asia, in the United States, and Israel as well. And using these labs, we collect the data from more than 100,000 people and up until now 250,000 hours of uh brain activity data. And that's the foundation of our model.

[00:03:16]>> So, I also understand you are uh going for FDA approval. Is there any word on this already?

[00:03:22]>> Yes, we actually met with the heads of CDRH, which this is the medical device uh unit of the FDA. We met them in person and did a live demo and presented our technology. We're now having formal written uh communication with them and we're getting getting ready for a submission uh sometime next year to submit our first product. It's important to note, these are observational products. They don't interfere with the human body. It's called class two. Um and this is software or AI as medical device. So, it's not the type of a regulatory process that have to do if you're developing say a drug or a surgical intervention that actually affects the body. This just measures the body and can't harm the body in any way.

[00:04:00]>> But let me ask you because at some point maybe when when this goes further and and a brain scan can become like a blood test, right? There are also a danger to this maybe that you can use it and you know exactly what you have even though you maybe before you didn't even know you would have you have anything?

[00:04:16]>> Uh if you're referring to the security aspect of it then you know, our company uh adheres to the highest tier of cybersecurity standards and we're treating brain data as as what it is which is very private. So everything is anonymized uh so in that respect uh it's very safe.

[00:04:36]>> And what about last question, but what about the the the the the helmet itself?

[00:04:41]At some point do you think it would expand beyond like mental conditions?

[00:04:45]You think something else can come up with your technology?

[00:04:49]>> So yes, uh what we do at the end of the day we're a deep learning and AI company, okay? And what we're doing and trying to do is to decode brain activity in general, not for specific use cases.

[00:05:01]The first thing we're addressing is brain health because we believe that's the most important thing to address now.

[00:05:07]Uh but we have other capabilities we haven't unveiled yet and we plan to unveil them uh as they mature.

[00:05:14]>> Well, you just secured a $52 million dollar fundraising round. So congratulations to you and thank you uh for being with me today uh CEO Hagai Lazar and CTO Gidi Grinstein. Thank you so much for being with me.

[00:05:27]>> Thank you. Thank you, Lauren.