Mathieu Galtier Details Morpheo's Novel Approach to Medical AI

Open-source, collaborative, and blockchain constrained, Morpheo is starting with sleep disorders and looking far beyond.

Rythm, makers of the Dreem sleep-monitoring headband, unveiled their AI project Morpheo in the spring. Rather than a massive, generalized machine-learning database built by a giant company, the idea is of a consortium-based approach that incorporates hospitals and startups and works to build the AI around a specific task from the start.

The first focus, which Morpheo project lead Mathieu Galtier calls its “proof of concept,” is to make diagnostic predictions about sleep. Galtier believes that in just a few years’ time, they will be able to make accurate predictions about what sleep conditions a patient will suffer from: sleep apnea, insomnia, or even depression that impacts sleep.

They’re able to get a wide variety of data, in no small part due to Rhythm’s headband, which Galtier says so far has already given them recordings of over 50,000 nights of sleep. Acquiring hospital data, and merging all of it together, is the difficult part. “Suppose I go to a hospital and say ‘give me your data, I’m going to do something great with it.’ Even if that’s true, the hospital is going to say in any case ‘no, we are not going to give you the data…there is a lot of information here, there is value in the data, and also these are sensitive data.’”

To counter that, and to create trust with hospitals and patients, the consortium emphasizes traceability and barriers that prevent developers from ever tying the data back to the patients.

“As a machine learner, I don’t look individually at the data, I look at the whole. What we are doing as a system where we can use these advanced analytics (machine learning, statistics) on the data without having access to the data…they remain in the scope of the hospital,” he said.

A big part of that effort is the use of blockchain to, in a sense, constrain the AI. “The hope is that these two technologies could cooperate, in some sense, and we may be among the first to bet at the same time on machine learning and blockchain technologies,” Galtier told Healthcare Analytics News.

“I think artificial intelligence is really promising in health, but still it needs to be orchestrated and conducted in a traceable, clean way. Blockchain is doing exactly that aspect. We’re not going to become the next bitcoin…the idea is to use blockchain, [to create] the transparency of logging all the applications in the public ledger,” he said. “The hope is that by providing trust in this aspect, the doors to new data are going to open and we’ll then get better algorithms in a sound environment.”

Compliance is another aspect of it: blockchain logged and entirely open source, Morpheo seems geared to avoid accusations of being a black box. Its algorithms can be validated. “You can say ‘if I use these data in this algorithm I get this performance,’” according to Galtier.

There may be a regional advantage to Rythm’s Morpheo, or perhaps just a friendly difference between the United States. While a lot of the groundbreaking work on AI for medicine is being done domestically, data siloing and institutional mistrust are even stronger in the US than they are in other parts of the world. Rythm is based in France, and received €5 million from the French government to start up.

“If you take Israel, for instance, they have all the data for all their patients completely gathered in the same data sets. This is extremely precious, in France we have something that is quite similar, it is not as scattered as in the United States,” Galtier said (he added, laughing, that France also has a lot of cheap engineers). He does see similarities in the way companies like to purport themselves, though.

“Latin cultures in Europe, we like to show off…I think the United States is in the same situation. We like this Watson approach, ‘we are about to solve cancer,’” he said. While narrative and confidence are important, he sees dangers in overpromising a platform’s capabilities.

Still, Galtier is quite confident in AI’s future for medicine. “For the average medical doctor, the artificial intelligence is going to come sooner than some expect. I strongly believe I will see it in my lifetime, a machine doing diagnosis and treatment, I think it’s quite obvious that we’ll see it earlier than later.”

Because academic institutions play a major role in the collaboration, Morpheo does place priority on the production of studies and academic papers from their work. They have already released a white paper, published all of their code, and submitted a paper for review regarding the Dreem headband. Galtier said to expect more studies to come in the near future. Ideally, he’d like to see enough people wearing the device, consistently feeding their systems more data to produce better and better predictive abilities.

Though sleep is Morpheo’s first frontier, Rythm is intent on applying its methodologies to the study of different conditions. Galtier says the wheels are already turning on two new consortiums: one for fertility and another for immunotherapy in oncology. The group is looking to begin fundraising for those soon.

Success, for Galtier, would mean the concepts at the core of Morpheo spreading across the healthcare AI landscape. “If we succeed…many different companies and startups working on different pathologies, but also big actors like Apple or Google, use this standout way of communicating and sharing data through this backend layer that makes the data more secure and hidden from developers,” he said. “Success in here for me is more of a broadcasting effect, where a lot of people use this technology.”