A 15-minute test: The Israeli development that will decode the brain?
A test lasting about a quarter of an hour may in the future provide doctors with objective information about brain activity and assist in the diagnosis of a long list of neurological and psychiatric disorders. The Israeli NeuroAI company Hemispheric is developing an artificial intelligence system designed to "decode" the brain's electrical signals—a field where diagnosis still relies in many cases on clinical assessment, questionnaires, and symptoms.
A test lasting about a quarter of an hour may in the future provide doctors with objective information about brain activity and assist in the diagnosis of a long list of neurological and psychiatric disorders. The Israeli NeuroAI company Hemispheric is developing an artificial intelligence system designed to "decode" the brain's electrical signals—a field where diagnosis still relies in many cases on clinical assessment, questionnaires, symptoms, and sometimes a lengthy process of trial and error.
The system, called Descartes, is based on the analysis of EEG data—a non-invasive test that measures electrical activity in the brain. During the test, the subject wears a cap equipped with sensors and performs tasks in front of a screen. The collected information is transferred to the artificial intelligence model, which analyzes the patterns of brain activity. The entire process is supposed to take about 15 minutes.
From depression to Alzheimer's
According to the company, the goal is to enable the identification and assessment of conditions including depression, anxiety, post-traumatic stress disorder (PTSD), attention deficit hyperactivity disorder (ADHD), obsessive-compulsive disorder, sleep disorders, Parkinson's, Alzheimer's, traumatic brain injury, and early cognitive decline. The company even aspires for the system to provide information on the severity of the condition.
Behind the system stands an artificial intelligence model with a scale of about six billion parameters. For its training, the company has established a system for collecting brain activity data over the years, building a database including information from about 100,000 subjects and approximately 250,000 hours of brain activity. The idea is similar to the way large language models learn to identify patterns within vast amounts of text: in this case, the model attempts to learn complex patterns within the brain's electrical activity and translate them into information that may have clinical significance.
Israeli soldiers also participated in a pilot
One of the applications being tested is PTSD diagnosis. The company conducted a pilot study among 122 Israeli soldiers who served in the Gaza Strip, with the aim of testing the system's ability to assess post-traumatic stress disorder. The company plans to submit results from a clinical trial in the field of PTSD to the FDA as early as next year, after the system has already been presented to the Center for Devices and Radiological Health of the US Food and Drug Administration.
The potential is great—but proof is still required
Alongside the promise, it is important to qualify: at the current stage, this is not a new blood test that one can request from a family doctor tomorrow, and the technology still needs to undergo clinical and regulatory validation. Experts quoted in publications about the company have warned that the technology has not yet been proven to the extent necessary for broad clinical use. The fact that the model was trained on a large database is not in itself proof that it can replace the accepted diagnostic process.
If the system passes clinical trials and regulatory requirements, the potential is significant: a move from diagnosis based in some diseases primarily on symptoms and clinical impression, to an additional tool that provides a quantitative measurement of brain activity. The big question now is whether artificial intelligence will indeed succeed in turning the brain's complex electrical signals into accurate medical information that doctors can rely on when making decisions.
