A Camera Combined With AI Could Help Further Diagnosis Of Movement Disorders Like Spinal Muscular Atrophy In Newborns

(MENAFN- The Conversation) Detecting impaired motor function as early as possible after birth with the help of a standard camera linked to intelligent systems can help in diagnosing infantile conditions characterised by hypotonia: reduction or loss of active muscle movement. A research team is investigating this approach based on the work carried out on babies with spinal muscular atrophy (SMA), a rare genetic neuromuscular disorder.

In 2019, the drug Zolgensma made headline news as the world's most expensive treatment. The price of this medicine for spinal muscular atrophy (SMA) has been set at nearly 2 million euros per injection.

And, at the same time, what an extraordinary breakthrough! It promises to provide a definitive cure for a genetic disease that, if survivors do live, leads to extremely severe disabilities in children affected by what is the world's first hereditary peripheral neurological disorder.

Behind these staggering figures lies an urgent medical situation. For these therapies to be effective, it is essential to identify the condition as early as possible. This is precisely the challenge our team set out to tackle, by using an ordinary camera coupled with artificial intelligence.

The aim was clear from the outset. Too many delays drastically reduce the effectiveness of the treatment. While the treatment can undeniably prevent neurons from dying, it cannot bring them back to life. Any delay in starting treatment – and, consequently, in diagnosis – represents an unacceptable loss of opportunity for patients.

What is spinal muscular atrophy?

SMA is a rare genetic condition that causes the progressive degeneration of the motor neurons that control muscle function.

In its most severe form – known as type 1 spinal muscular atrophy, infants with the condition rapidly lose the ability to move, sit up and eventually breathe. Without treatment, life expectancy rarely exceeds two years. In France, there is one case in every 6,000 to 10,000 births.

Since the introduction of new treatments, the prognosis has changed dramatically for children treated at a very early stage. Spinal muscular atrophy has, in fact, been included in routine neonatal screening in France since 2025. Not all countries have this screening in place yet. And even where it does exist, a complementary tool for rapid clinical assessment remains highly useful.

Hypotonia refers to a reduction in, or even the complete absence of, active movements. This symptom is non-specific and can be associated with many other childhood conditions, but it remains the first indicator, the one that allows the diagnosis to be raised promptly.

The challenge with clinical diagnosis

Before genetic testing, diagnosis relies on clinical observation. A specialist examines the newborn's muscle tone and reflexes. The term“hypotonic infant” is often used: the baby appears limp, and their limbs droop without resistance.

However, this assessment is subjective, as it varies from one practitioner to another, and is all too often made at a very late stage, by which time the symptoms have already progressed significantly.

This, then, is the real challenge: the therapeutic window is narrow.

Our approach: motion capture

Our study involved 25 infants hospitalised in paediatric resuscitation units: 5 cases with genetically confirmed SMA and 20 patients with normal neurological results. We used computer vision to analyse infants' spontaneous movements. The principle is simple: the infant is placed on a plain background, while a conventional camera captures its movements for sixty seconds; an artificial intelligence algorithm then analyses the video footage frame by frame.

In practical terms, this study is based on a 3-step video analysis pipeline, as shown in Figure 1.

The system first reconstructs a“digital skeleton” of the infant based on twelve anatomical joints, eight limb segments, and four movement angles (pose estimations), using a real-time human pose estimation method called Alpha Pose.

From this animated skeleton, hundreds of parameters are calculated (amplitude of gestures, movement limitations (depth), symmetry, frequency...). A total of 108 features were extracted. We then trained a supervised learning algorithm (of the type XGBoost ) capable of distinguishing typical motor skills from 'altered' motor skills characteristic to spinal muscular atrophy.

Promising results

The results are encouraging: the algorithm correctly classified both groups with an accuracy of 97%.

At the top of the most discriminant parameters, we find the depth of movements or movement limitation; in other words the ability of the infant to move its limbs in space. Babies with SMA show a significant difference in depth-axis motor motion with detection sensitivity greater than 97%. The tool we developed rigorously measures this, turning a visual impression into objective data. What the clinician's eye intuitively perceives, the algorithm quantifies with precision.

To make the AI outcome relatable, and thus incorporate it in clinical practice, we used a mathematical method called Shapley Additive Explanations (SHAP), which allows you to view the parameters that have been most influential in each algorithm decision.

A tool for other diseases causing hypotonia in infants

Our study was conducted before 2025, at a time when no systematic screening existed in France. Since then, SMA has been integrated into France's national newborn screening program, which retrospectively confirms the urgency behind our study.

But beyond SMA, many other diseases cause hypotonia in infants without any rapid assessment tool. Our work continues in this direction. While this tool does not set out to replace doctors, in a few minutes it gives practitioners an initial objective pointer to guide their diagnosis without specialised equipment.

Artificial intelligence does not perform miracles, but it renders what is invisible to the naked eye visible. When each week counts for an infant, AI's contribution to diagnosis becomes valuable.

The Axa science philanthropy is now part of the Axa Foundation for Human Progress, which brings together the commitments of Axa Group and Mutuelles d'Assurances in the fields of Science, Nature, Solidarity, and Culture. Before 2025, the global science philanthropy was held by the Axa Research Fund, which has supported over 750 projects around the world since its inception back in 2007. To learn more, visit Axa Foundation for Human Progress.

A weekly e-mail in English featuring expertise from scholars and researchers. It provides an introduction to the diversity of research coming out of the continent and considers some of the key issues facing European countries. Get the newsletter!

This article was originally published in French

MENAFN06102026000199003603ID1111768690