Sonio, the AI ​​made in France which improves prenatal diagnosis

“See that stack of books? There are over 400 prenatal syndromes referenced in there

Sonio, the AI ​​made in France which improves prenatal diagnosis

“See that stack of books? There are over 400 prenatal syndromes referenced in there. The problem is that all this doesn't fit into a doctor's brain, even the best one! » says Professor Yves Ville, head of the maternity department at the Necker-Enfants Malades hospital in Paris, pointing to an impressive column of obstetric medicine books that runs from the floor to the ceiling of his office. “A specialist in neonatology stores, at best, around fifty of the most common pathologies. He needs an augmented brain to manage this mass of information. »

Objective (almost) achieved. Sonio, an artificial intelligence system specializing in the prenatal health of children and pregnant women, which he co-founded, is a sort of extension of his brain and that of other practitioners. Born in 2020 from the collaboration between Necker hospital and the Polytechnic school, Sonio provides a control solution to obstetrician-gynecologists, general practitioners and sonographers during pregnancy monitoring and allows them to better detect potential anomalies in utero.

Sonio is not his first attempt. The company began by developing Sonio Diagnostics, already widely distributed in France. “Diagnostics is used in half of the Multidisciplinary Prenatal Diagnostic Centers [CPDPN] in France,” says David Amouyal, COO [director of operations] and co-founder of Sonio. This is a diagnostic aid which, thanks to an intelligent decision tree, guides the practitioner towards potential anomalies, and therefore towards the correct diagnosis. »

In developmental medicine, the devil is in the details. Identifying a problem in the fetus requires doctors to “assemble” clinical signs which, taken in isolation, are often banal. “Our head is made up of somewhat disparate elements which, when we put them together, can sign what we call syndromes. An anomaly observed in the images can lead to another, explains Professor Ville. For example, if I see a club foot in the image [a congenital malformation of the lower limb, Editor's note], I can report it to Sonio who suggests that I investigate further. Are there ventricles in the brain that are a little large? Is the spine normal? Taken together, these signs can suggest a neurological problem: a lack of closure of the spine, called spina bifida. »

Sometimes, the most serious pathologies are hidden behind weak signals. This is the case, for example, of burden syndrome, associated with serious mental retardation. “In this picture, there are malformations of the genitals, the heart, the brain,” continues the doctor. But each sign can be so subtle that in three quarters of cases, the diagnosis is not made and the problem is only noticed at birth. Sonio will tick and facilitate the diagnosis. »

Fetal malformations often have a so-called ontological link: either they have a common origin, or one leads to the other. Kinds of embryology equations that mathematical artificial intelligence can help solve. Sonio therefore allows you to approach an emotionally charged exam more calmly. “When the operator “stumbles” upon something unusual, or frankly abnormal, he may feel anxious, not knowing what to do. Anxiety that the pregnant woman feels herself. Having a protocol to follow when discovering an anomaly avoids emotional reactions,” assures the obstetrician.

It is with its second born that the French company is now making an American breakthrough: Sonio Detect. Detect also helps the operator not to “miss” anything during the ultrasound examination. But it is positioned differently – and in addition to Diagnostics – by monitoring the quality of images of the fetal organs. “An ultrasound examination is a very long checklist,” explains Professor Ville. We're talking about sixty cutting plans, with eight to ten things to check on each one. Which makes around 600 points to check. This requires lucidity and constant attention to be able to honestly say that everyone has been seen. »

In fact, things frequently escape the attention of healthcare professionals: half of malformations – which affect 5% of babies – are only noticed at birth. “A significant proportion of ultrasound examinations are carried out with quality standards lower than those recommended,” notes the head of the Necker maternity ward. You can take a cross-sectional image of the brain, but you can take it wrong! In this case, the risk of not detecting an anomaly increases. » Hence the valuable help of AI: Sonio Detect has proven its performance on more than 17,000 ultrasound images, with a sensitivity of more than 92% to the quality of these images.

Detect is all the more useful across the Atlantic where the sonographers – who perform the ultrasound – are not doctors. “He must take around a hundred images in a standard second trimester examination,” reports David Amouyal. It's a real sprint, which they must run in around twenty minutes. And we're talking about 10 to 15 patients per day. »

And it’s better for technicians not to get their feet wet. “The quality of the images depends on their reading, subsequently by the obstetrician,” he continues. In addition, things can become more complex and give rise to lawsuits. In France, things are different and there is not this level of separation between the sonographer and the practitioner, but ultrasound scans are increasingly carried out by midwives. »

If Sonio had an interesting opportunity in the United States, its founders aim to have Detect approved in Europe. They even want to go further: to artificial intelligence that would identify suspected anomalies. “Once the image is well taken, that it meets the quality criteria, we want to try to make the machine identify an abnormal positioning, or size of structure, for example,” says David Amouyal. Today, this remains at the doctor's complete discretion. »