Characterization by Automated System on Infantile Spasmes

Infantile spasms are defined by the occurrence of epileptic episodes characterized by the appearance of very specific motor seizures, made up of rapid, repeated contractions in flexion or extension known as spasms. This syndrome is of high concern as it will lead to mental retardation if it is not early identified and treated. Most often, spasms are characterized by sudden contractions of the body in flexion. More rarely, spasms occur in extension: the neck and legs suddenly tense, and the arms move upwards or sideways. Spasms may be accompanied by eye revulsion. They are brief, lasting 0.5 to 2 seconds each. At the start of the attack, they may be isolated, but then may occur in series lasting up to several tens of minutes.

Because of its nature and duration, infantile spasm is often difficult to identify and is often confused with benign motor manifestations in children, such as hiccups or muscular jerks. This difficulty often leads to delays in diagnosis. Indeed, patients are often seen at the stage when cognitive regression has set in.

The gold standard for diagnosing infantile spasms is to capture them on video-EEG to confirm the ictal correlate of the seizure. Although multiple variations are known, the ictal correlation is often a diffuse slow wave of high amplitude with subsequent electrodecrement. Yet continuous video electroencephalogram is an expensive and time-consuming resource often unavailable in developing countries and in many parts of the developed world. Furthermore, in an appropriate clinical scenario, continuous video-EEG may be unnecessary. For example, a home video examination with typical spasms combined with a definitively epileptic ambulatory EEG may certainly be adequate for the diagnosis of infantile spasms

As part of Mr. Diop's thesis, the study team have developed a system for analyzing videos acquired in 2 dimensions from a simple smartphone or webcam and highlighting the existence of spasms in a child. The principle is to use computer vision and computer learning model to identify the spasms from these videos. The first prototype of this system achieved a positive predictive value of 77%, which is very good considering the small sample used (< 100) but quite insufficient to obtain a diagnostic prediction for medical use, for which we hope for sensitivity and specificity of around 95%.

The aim is to achieve a sensitivity and specificity of over 95% so that we can offer this detection system to healthcare professionals and parents of children who do not have rapid access to diagnosis. The aim is to develop a system enabling broad screening at the population level so that identified children can be more rapidly directed towards the healthcare system and appropriate treatment for their disease.

To achieve this goal, we need access to a standardized reference database that currently exists in the various pediatric electrophysiology laboratories of the Assistance Publique - Hôpitaux de Paris and the Île-de-France region. we propose using the films stored in these laboratories to teach the computer to recognize spasms that have already been formally identified by electrophysiologists and by ictal concordance.