Imagine a future where diagnosing cognitive disorders like Alzheimer’s or PTSD is as straightforward as a blood test. This ambitious vision is at the heart of Hemispheric, a groundbreaking startup co-founded by Gidi Littwin, an inventor behind Apple’s FaceID and Vision Pro technologies. After six years of intensive development, Hemispheric has successfully raised $52 million in funding to bring its frontier AI model closer to reality.
Littwin’s journey began after leaving Apple in 2020, seeking a new challenge. It was a cold message on LinkedIn from Hagai Lalazar, Hemispheric’s other co-founder, that sparked the collaboration. Lalazar had already started developing AI to study the brain non-invasively, and he recognized Littwin’s commercial acumen and experience in building massive data collection operations.
From Apple’s AR to Brain AI
Littwin’s tenure at Apple involved pioneering work on FaceID and hand-tracking for augmented reality, requiring him to collect “hundreds of thousands of subjects’ worth of data.” This experience proved invaluable for Hemispheric’s ambitious goals. He understood that a similar, massive data collection effort would be essential to train their deep-learning models effectively.
Historically, diagnosing conditions like depression, Alzheimer’s, and Parkinson’s has relied heavily on subjective questionnaires and behavioral observations. This is largely because each individual’s brain activity presents uniquely, making objective analysis challenging. Hemispheric’s solution was to gather what Littwin calls their “most prized possession”: an unprecedented dataset.
They collected a quarter of a million hours of brain data from 100,000 paid volunteers across Asia, Tel Aviv, and Boston. These participants engaged in game-like activities designed to activate specific brain regions, providing a rich, diverse pool of information. This vast dataset became the foundation for training their revolutionary AI model.
Decoding the Brain with Frontier AI
Hemispheric’s frontier model is designed to infer brain function from electrical activity within the skull. This process is analogous to how large language models statistically analyze text to deduce meaning. The team then rigorously tested their generalized model on subsets of individuals already diagnosed with conditions like PTSD, schizophrenia, and depression, reporting accurate deductions about their brain health.
A significant focus for the team is a current clinical study to determine if their model can accurately diagnose and even predict Alzheimer’s disease. This research holds immense promise for early intervention and improved patient outcomes. The first product, targeting PTSD, is slated for FDA submission early next year, with a public rollout hoped for in late 2027.
A Blood Test for the Brain
The diagnostic process envisioned by Hemispheric is remarkably simple and non-invasive. A patient would wear a lightweight EEG headset for about 15 minutes, interacting with an app on a tablet to stimulate specific brain activity. Hemispheric’s AI model then helps clinicians decode these electrical signals, enabling accurate diagnoses, predicting the most effective treatment interventions, and monitoring progress over time.
Hagai Lalazar envisions a future where this technology is as common and accessible as a blood test. “The device is going to be very, very cheap,” he says, “it will be able to be sold and distributed throughout mental health clinics, hospitals, and even psychologists’ offices.” This widespread availability could democratize access to advanced neurological diagnostics, much like AI-assisted tools are already accelerating lung cancer diagnoses in Europe.
Investing in the Future of Brain Health
Hemispheric has secured early-stage funding from prominent American and Israeli venture capital firms, as well as individual investors, including early Uber-backer Howard Morgan. This substantial investment will fuel several critical initiatives:
- Advancing partnerships with governments, healthcare organizations, and pharmaceutical firms.
- Expanding their team with new hires in the US.
- Working diligently toward crucial regulatory approvals.
- Collecting even more brain data from millions of people to continually improve and refine their AI models.
Furthermore, the team isn’t just relying on existing technology; they are actively developing their own advanced brain scanners. Littwin notes that traditional EEGs “were never built for machine learning and definitely not deep learning.” Their proprietary scanners aim to capture richer, more useful data specifically optimized for their sophisticated AI models, pushing the boundaries of what’s possible in brain health diagnostics.
Source: Wired – AI
