Israeli chatbot can diagnose early stages of Alzheimer's disease

The chatbot, which is still in the testing phase, is based on the understanding that before affecting memory, Alzheimer's disease negatively impacts the so-called brain orientation system.

“Hundreds of drugs have been developed for the treatment of the most common form of dementia,” says Dr. Shahar Arzi, director of the psychoneurological laboratory at the clinic. “Want to know how many of them have been recognized as effective? Zero.”

But if the disease could be detected at preclinical stages, perhaps some of the new biological agents showing impressive results in other areas of neurology could be effective in the early stages of Alzheimer's disease.

Together with colleagues, Dr. Arzi developed a computer system to identify early signs of dementia. The system, named “Clara” (a nod to 'clarity of mind'), is an AI-based chatbot that asks patients questions about themselves, their relationships with people, places, and events. It then uses machine learning to compare the information obtained with baseline data and creates a computer test tailored to the individual that can diagnose early-stage Alzheimer's disease.

According to the results of a study published by the team in the Proceedings of the National Academy of Sciences and the scientific journal Neuropsychology published by the American Psychological Association, the accuracy of the new method reaches 95%. In developing “Clara,” specialists relied on relatively recent data about brain function and what happens to it during Alzheimer's disease.

Dementia affects the brain's “orientation system,” which determines how a person interacts with the external world. “Testing memory is easy,” says Arzi. “I will give you three words and then after a while ask you to repeat them. But testing orientation is quite difficult.” For example, a patient may remember the assassination of President John F. Kennedy and the election of Barack Obama but struggle to say which of these events occurred first. Or they may recognize their husband (wife) and their treating physician but cannot say who is closer to them.

Orientation can be assessed using functional magnetic resonance imaging. When you see a picture of your daughter, your brain will react quite differently than when seeing a picture of someone else's child. “The intersection between how the person relates to the world around them and the brain mechanisms disrupted by Alzheimer's disease is striking,” says Dr. Arzi.

According to him, at preclinical stages of dementia, the orientation system begins to function worse, but people can still compensate for these impairments by using other resources, such as memory. “They can write themselves reminder notes, for example. And although it takes more effort, a person is still able to achieve the desired result,” says the doctor.

Only when both systems – orientation and memory – cross a certain threshold does the disease become apparent. Only then do people seek help, but it is already too late for effective treatment.

The main feature of “Clara” is that it asks questions based on the patient's personal orientation system; these are not general questions related to memory and other cognitive functions.

How the technology was created

Dr. Arzi's initial idea was to launch a chatbot and gather information from Facebook and other social networks: “This is what we worked on at the very beginning, spending two years writing the corresponding code.” And when everything was ready, the Cambridge Analytica scandal broke out.

Cambridge Analytica was a consulting firm that collected and analyzed publicly available data from Facebook users to influence voting in the 2016 U.S. elections. In response, Facebook disabled the feature that allowed third parties to access user data. This story forced Arzi and his team to start from scratch.

“This was one of those cases where they say: if it weren't for bad luck, there would be no happiness,” recalls the doctor. “The information we were getting from Facebook was not ideal, and the decision to use artificial intelligence turned out to be much more advantageous.”

The authors suggest that “Clara” will be free and available through medical offices and by installing the corresponding application on mobile devices and computers. The pilot Android and web versions currently support only English, Hebrew, Chinese, and Portuguese. Versions in French, Russian, Arabic, and Japanese are currently in development.

“For now, ‘Clara’ is not yet ready for ‘release into the world.’ We are still testing the effectiveness of the intelligent agent used,” notes Arzi. The second year of a five-year study at Harvard is currently underway, which is supposed to compare the data collected by the system with data on Alzheimer's disease markers obtained through positron emission tomography, quantitative and functional MRI, and other neuropsychological tests.

Over the past year, the Tel Aviv medical center Assuta has been conducting its own trials of “Clara” using PET and functional MRI.

The security of the data collected by the chatbot has been ensured by the Israeli company Guardicore, specializing in cybersecurity. “When we contacted them, we insisted on one condition: that only volunteers work on the project,” recalls Arzi. “I went to the company's office and gave a lecture attended by a hundred people. At the end, I asked who would like to volunteer – all one hundred raised their hands.”

The spirit of volunteering permeates the entire story of “Clara.” Once, Arzi traveled to Brazil to demonstrate his system. There he met a volunteer who offered to translate the chatbot into Portuguese. He said, “I can't bring all patients to the best doctors, but I can buy tablets, bring them to villages, and teach people how to use them.”