One of the aims of both political and medical authorities is to diagnose neurodevelopmental syndromes such as autism as early as possible. Behavioural treatments are all the more effective if they start early, during the pivotal period of plasticity and learning in the first 2-3 years of life. In France, as in most other European countries, autism is usually diagnosed around the age of 5-6.
Based on the fact that autism is ‘born’ in utero, we tested the hypothesis that analysis of maternity data could make it possible to identify babies with autism from birth. Using a Machine Learning (ML) approach, we identified autistic and neurotypical children born in the same maternity hospital and analysed all the maternity hospital data retrospectively. It should be noted that in France, over 150 parameters are normally routinely collected, which makes for a large data set. Our measurements have enabled us to identify virtually all babies born with neurotypical features and almost half of those born with autism, with a high degree of specificity and few false positives. It should be noted from the outset that this prognosis is only valid at birth, as in utero parameters are not sufficient, in line with the major role played by the conditions of birth; this rules out any hint of eugenics, no, birth data must be included for the programme to work.
When we analyse the parameters that led the programme to identify ‘autistic’ babies, some were expected (viral or bacterial inflammation, CMV, etc.), others were not (femur size during the 2nd trimester of pregnancy, etc.). Others are very interesting, even if they are not easily understood, such as the early head-down presentation in those who go on to become autistic. The most interesting element in terms of developmental biology is the size of the perimeter of the encephalon, which is larger in the autistic foetus than in the neurotypical. In fact, just before birth, the brains appear to be larger. Data in mice show that neurons and brain structures such as the hippocampus continue to grow during birth – by comparing size a few hours before and after birth, which is of course impossible to do in women. Neurons grow in size during birth in autistic mice and not in control mice. Moreover, US data shows that the brains of children and adolescents are larger in autistic than in neurotypical mice. In other words, during maturation in utero, there is accelerated brain growth. We are continuing this work, firstly by validating it in other maternity units, with the ultimate aim of making France the1st country in the world with such an early prognosis.
At the same time, drawing on our expertise in ML, we have begun to use this approach in a number of different ways. Firstly, to re-analyse the data from the failed phase 3 trial (see other article), but also to examine preoperative functional imaging in Glioblastoma surgery and other brain tumours to see whether ML could establish a rapid prognosis of the pathology to be operated on. A whole range of possibilities is being examined by our start-up – backed by 2 ML experts – H Rabiei and M Begnis – to open up the field of possibilities in which we can help analyse large datasets to establish a prognosis or analyse data for a clinical trial.
For further information : contact@babiomedical.com
