Machine learning to improve neonates' General Movements Assessment – MAGMA

From birth, the movement of the baby as a whole and, in particular, of each segment (arms, legs, trunk) carries essential information whose expression depends on the baby's neuro-anatomical and functional cerebral integrity. However, it has been neglected as an early predictive marker of later development, mainly because of three elements: an absence of large banks of recordings integrating 3D videos allowing an analysis of the movement of the individual in the three planes of space, unsuitable analysis tools, and the lack of interaction between the medical world and that of signal processing and artificial intelligence scientists. We can remove these obstacles thanks to the synergy of two teams from the Jean Monnet University of Saint-Etienne, already working in collaboration and an association supporting pediatric research.

General movements represent a spontaneous motor activity occurring in the fetus from nine weeks of amenorrhea (SA) and continuing after birth until the appearance of controlled movements, around four months of corrected age. General Movement Assessment (GMA) is a validated assessment of brain maturation based on qualitative film analysis of general movements' complexity, variability, and fluidity. Reliable and reproducible, this observational method can identify very early, i.e., before discharge from the hospital for premature, children who are at risk of abnormal neurodevelopmental trajectories.
Identifying such a risk is essential to initiate as early as possible a personalized intervention (psychomotricity, motor physiotherapy, postural care, etc.) aimed at improving these children’s motor and cognitive development. However, GMA analysis is very time-consuming because it requires viewing long filmed sequences of the baby to select the motor parameters of interest. This analysis also requires significant training on the part of the clinician so that the qualitative criteria of these movements can be of high predictive value and are fully exploited. All these elements limit general movement assessment in everyday practice within neonatology units.

By relying on a consortium of researchers from Saint-Etienne from 2 labeled research teams from the Jean Monnet University of Saint-Etienne (Lab. SAINBIOSE and Lab. Hubert CURIEN) and the ADERPS Association (law 1901), hosted at the CHU of Saint-Etienne whose object is to work in pediatric research and the neuro-developmental follow-up of vulnerable children, we propose to design an early predictive tool for the development of premature or at-risk children, based on automation analysis of general movements by deep learning.

Two stages are scheduled to achieve this objective: the MAGMA EXPLORER study and the MAGMA PROGNOSIS-0-2 study.

- The MAGMA EXPLORER study will develop a method of automated analysis by artificial intelligence of the general movements produced from birth.
- The MAGMA PROGNOSIS 0-2 study will model the correlations between the characteristics of these general movements at birth (complexity, fluidity, variability) and their evolutionary data at two years (BAYLEY-4 scale)

Through an unprecedented approach to the care of newborns at high neuro-developmental risk, the MAGMA project aims to offer the first international reference method for aiding in the prognosis of the ultra-early development of children at high neuro-developmental risk.