The AI catching what MRIs keep missing

Some of the most important work in AI right now is happening in healthcare. Women’s health has been under-researched and under-funded for generations, particularly in relation to endometriosis.

Élise Mekkaoui and Raphaelle Taub refused to accept that as the way things are. One had lived it, the other had the scientific training to do something about it.

Their story is a reminder that the best AI companies often start with a problem someone could not bear to leave alone, and that is exactly why they belong in our 2026 class of the Most Interesting Women in AI.

-MV

What’s in store:

Endometriosis has nowhere to hide in the AI era.
Your Weekly AI Roundup!
Poll of the week: Where AI still falls short?
Get our free Claude at work guide!

Read Time: 6 minutes

WOMEN IN AI 2026

The Women Who Decided Seven Years Was Long Enough

For four years, Élise Mekkaoui went from doctor to doctor, trying to get a definitive answer about the pain she was living with.

She got a lot of other answers instead: That it was manageable, that she was probably fine. At one point she was told the most reliable way to confirm what was wrong was surgery.

"If someone told me to do a surgery to get diagnosed for endometriosis," she says now, "I would've said, okay, no, maybe I'm just going to stay in pain for my entire life."

She did not stay in pain for her entire life. She co-founded a company instead.

Raphaelle Taub was a physicist who loved the lab and hated her job in reinsurance. She had a PhD in fundamental mechanics, had studied at some of France's most elite academic institutions, and had recently begun to feel that calculating catastrophic risk for insurance companies was not quite what her brain was designed for.

She and Élise were friends, and one conversation about what Élise had been living through was enough. Raphaelle went home and wrote in her journal that the idea was "spiralling out of control." She meant it as a note to herself. It became the plan anyway.

If someone told me to do a surgery to get diagnosed for endometriosis, I would've said, okay, no, maybe I'm just going to stay in pain for my entire life.

Endometriosis affects one in ten women globally. The average time from first symptom to diagnosis is seven years. 3 out of 4 cases are misdiagnosed at some point, and even when a doctor orders an MRI, up to 60% of cases are still missed. The disease is not hiding. The tools for finding it simply haven't been built.

Nobody built them until Élise and Raphaelle founded Matricis.

EndomAI sits inside the radiology workflow. The tech was trained on the world's largest dataset of endometriosis MRIs: 14,000 unique patients sourced across three continents, labelled at the pixel level by a team of expert radiologists. A radiologist reads the scan and EndomAI reads it alongside them, recognising the patterns that otherwise go undetected.

Raphaelle is CEO and Élise is COO, they built Matricis at INRIA and Agoranov, two of France's leading deep tech incubators.

The reception from the medical community surprised Raphaelle. "Honestly, it was pretty good," she says. "Endometriosis is really painful to read because it can be spread around the whole pelvis area. It's quite long to do. It's quite stressful."

Radiologists, she found, already know the training for endometriosis is not good enough. They are not looking for a fight. They are looking for help.

It took 16 years, but imaging finally moved from the sidelines to the center of endometriosis diagnosis.

In 2026, the American College of Obstetricians and Gynecologists updated its clinical guidelines on endometriosis diagnosis for the first time in sixteen years. Imaging moved to the centre of how the condition should be identified. Élise posted about it on Linkedin with four exclamation marks.

"It took 16 years, but imaging finally moved from the sidelines to the center of endometriosis diagnosis." She was happy, she says, mainly for the patients.

A clinical pilot with SimonMed launched in April 2026, using EndomAI, the first AI-assisted MRI workflow designed specifically for endometriosis, now operating in a real clinical setting with real patients. FDA clearance is planned for Q4 2027. Matricis has been supported by BPI France and région Île-de-France, and is a France 2030 laureate.

Endometrosis's seven-year diagnosis delay is not a fixed fact. It is a consequence of under-investment and a medical system that moved too slowly for too long. Elise and Raphaelle started because they couldn't accept that as the answer.

EndomAI is already in clinical use. Find out more at matricis.ai.

Get our free Claude at work guide below!

Get The Claude Guide →

POLL OF THE WEEK

What do you wish AI was better at?

Remembering context

Being concise

Knowing when to stop

Other - tell us!

Full results will be at the bottom of tomorrow’s newsletter!

THIS WEEK IN AI

Reddit locks the door, a satellite writes poetry, and dolphins have a translator

Reddit is kicking out the lurkers. Google’s AI satellite sent back a haiku from orbit. And researchers are using AI to decode dolphin communication, which Maria correctly identifies as the best joke of the week.

Also Anthropic warned Claude might soon give birth to itself. Cool!

Here’s what mattered:

Reddit announced it will require users to log in to view most posts, as search referrals fell by over 1 million per week in October, with the platform betting a smaller but registered audience is worth more to advertisers than anonymous drive-by traffic.
Prime Video debuted Castle Walls, the first fully AI-created series on a major streaming platform, produced by Turkish studio Ay Yapım across two 17-minute episodes, with reviewers describing characters as resembling melting Madame Tussauds statues.
Google successfully ran an AI model on its Project Suncatcher satellite in orbit, with the onboard chips generating a three-line haiku about space and transmitting it back to Earth, confirming AI computing can function outside the atmosphere.
DeepSeek launched Creator Mode, letting users build custom AI plugins through written instructions without writing code, alongside an improved coding assistant, as the company raised around $15 billion in its latest funding round.
The UK’s National Cyber Security Centre warned that AI-powered ransomware attacks on hospitals, airports and power stations are highly likely by 2028, with Russia and China expected to deploy AI cyberattacks regularly in the coming months.
Anthropic warned in updated financial filings that its models could soon help build more powerful versions of themselves through recursive self-improvement, admitting it has previously missed major capability jumps in its own models until after they happened.
Anthropic published FlashProve, an open-source tool using mathematical proof to verify which AI model produced a given response, designed to stop companies from secretly swapping expensive models for cheaper ones without users knowing.
Scientists in Germany filmed skyrmions, tiny magnetic structures, breaking apart in real time for the first time, opening a path toward computers that process information using magnetism rather than electricity, potentially matching the energy efficiency of the human brain.
Researchers at Georgia Tech and Google DeepMind are working with dolphin communication pioneer Denise Herzing to use AI to analyse 40 years of recordings and identify patterns, while testing wearable computers that play dolphin sounds to see how the animals respond.
OpenAI launched Images 2.6 across ChatGPT, Codex and the API, introducing GIF creation, SVG exports and two generation modes, with the faster Flare model producing images in as little as six seconds as users now generate four billion images per week.