AWS Pioneers Project
European innovation, told by those who built it
Callyope develops AI to help predict mental health crises
When Martin Denais, now CEO and co-founder of healthtech start-up Callyope, reconnected with his friend Rachid Riad, he was already considering a different career path.
"After seven years in finance, I was looking for a new career path where I could have more impact, and be more useful for society," he explains.
Meet Martin Denais
CEO and co-founder of Callyope
Europe's mental health crisis
Rachid, whom Denais met at engineering school, was finishing his PhD working on AI applied to neuroscience. He had developed speech analysis technology that could monitor symptoms in Huntington's and Parkinson's diseases. The pair, along with third co-founder Xuan-Nga, had the idea of applying the technology to mental health.
"We were struck to see how unequipped mental health clinicians are. Mental health is the only discipline in healthcare where most medical decisions are made with no objective date to guide care, like making a diagnosis or prescribing a treatment are all based on the conversation between clinicians and their patients," says Denais.
"Time with patients during consultations can be constrained and infrequent so it’s really a matter of providing more objective data to the clinician, more often."
Europe is in the midst of a mental health crisis, spending around 4% of its GDP—approximately 600 billion euros—on mental health. In France alone, there are an estimated 600,000 people with schizophrenia, more than 1 million people with bipolar disorder, and more than a million experiencing chronic depression.
On average, there is a 50% rehospitalisation rate within 12 months of a patient experiencing a mental health crisis, and 90% in a lifetime.
"Each relapse contributes to unemployment. Patients with schizophrenia who relapse three times are 10 times more likely to face lifelong unemployment,” says Denais.
Callyope is on a mission to change that narrative by providing mental health clinicians with more frequent and objective clinical assessments, so that crisis can be avoided.
Understanding speech patterns helps
"If you want to diagnose schizophrenia, there are five key symptoms, and one of them is discourse disorganisation. With bipolar disorders, it's the length of sentences and the speech rate that fluctuates. Someone with mania is going to speak really, really quickly—pressured speech, super long sentences—and it tends to be the opposite when people are depressed,” says Denais.
The firm developed its own AI foundation model that can assess patient speech related to mental health symptoms and anticipate relapses.
"We've proven that we only need 30 seconds of speech to assess symptoms like psychosis, anxiety, depression, and cognitive decline," Denais says.
This means clinicians can monitor their patients remotely. From the patient's point of view, the process is simple.
"The patients download our app and then they decide when and how they want to be monitored. I think that's super important to make these patients feel that they are part of the pathway and have an active role," says Denais.
"They can keep a voice journal on our app. They are prompted once a week to talk about their day and how they have been sleeping, and we can analyse this conversation."
Patients can also select trusted caregivers—parents, siblings, or friends— that can call them through the Callyope app.
Clinical validation and regulatory approval
The app is currently being trialled in several French hospitals. Callyope is running eight clinical trials with 1,000 patients to prove it can assess symptoms accurately.
"You have to have this thorough clinical validation before you can put the device in the hands of patients," explains Denais.
Callyope’s model was trained on largest speech dataset in neuroscience totalling 20,000 individual speakers, suffering from 10 different brain illnesses and speaking 7 different languages.
AWS opens doors
"Amazon Web Services (AWS) has proved invaluable, and Callyope is scaling its training workloads and experiments on AWS infrastructure," says Denais.
Beyond the tech, AWS is also proving to have a useful network. "You have a lot of regulatory hurdles. Before you do anything with patient data, it takes a lot of time to write protocols, to go through ethics committees, to comply with all the rules around handling patient data. You have to convince the doctors to use it, the patients to use the app, and potential investors to finance our company. Beyond its technology expertise, AWS has helped us build our business and get the right advice," he explains.
What's next
"We just launched our platform (also named Callyope), so at the moment we're very focused on selling that to psychiatric hospitals, implementing it, and entering the daily lives of clinicians," explains Denais.
"And we are pushing our trials on the remote patient monitoring aspect. We're about to launch a few more research collaborations which will be announced soon. So more partnerships, more data, more external validation and more clinical trust." Ultimately, the goal is to become the “Operating System”, or “super-assistant” of mental health clinicians."
"In a few months' time, we are going to be making the lives of mental health clinicians a lot easier by saving them time and essentially accessing more easily the patient data that is buried in pages and pages of PDF documents at the moment," he says.
"The current mental health pathway is just not working. With productivity tools and better monitoring technologies, we can do a huge job in improving medical outcomes at scale in psychiatry."
Behind the scenes