Autism, speech therapists and algorithms

Changed on 15/01/2021

An unusual experience took place by videoconference a few days before Christmas with the Stars team: an event bringing together student speech therapists, clinicians and team members as part of a research project to create algorithms to help diagnose people with autism. Interview with Susanne Thummler, lecturer and hospital practitioner at Université Côte d'Azur and the CHU-Lenval of Nice, who has been a member of the Stars team since Septembre 2020.

A project on autism

The Inria Stars team is part of a multidisciplinary project led by Séverine Dubuisson, from the LIS (Computer Science and Systems Laboratory) at the University of Aix Marseille, where everyone has a specific role to play.

5 partners united around a project.

LIS focuses on the interaction of the gaze
The development part of CobTek and the Centre Ressources Autisme (CRA) focuse on the clinical part and provides the interview videos.
The company Wita, which specialises in 3D videos, is in charge of the installation, software, etc.
Inria's Stars team for the analysis part of the general movement of the body

Video motion analysis, one of the Stars team's research topics, is an important brick in neurodevelopment, particularly for people with a diagnosis of autism, as it allows the patient to be assessed in a non-invasive manner.

Indeed, the tool used, and internationally known as "ADOS" (Autism Diagnostic Observation Schedule), although it allows a filmed evaluation of 30-60mn according to a standardised procedure, does not allow clinicians to make a diagnosis during the session.
An in-depth study by multiple viewing of the same video is necessary in order to be able to codify them and evaluate the session accurately.

Speech and language therapists "listening to the gaze".

While the link between the autism research project and speech and language therapists is not obvious at first glance, it becomes clearer when Suzanne Thummler explains how their particular skills and qualities give them the opportunity to make an important contribution to this study.

Indeed, they are people who, by their profession, have a great sense of observation, an indispensable and unavoidable element in order to identify each and every detail of the patient's behaviour.

It goes without saying that everything has been put in place for this study to ensure that the ethical rules are respected.

Research training, an important part of the course

Speech and language therapy studies are now associated with the university (bachelor's then master's degree) and take place over 5 years (instead of 4) with access to research training during the two years of the master's degree.
It is within this framework that speech and language therapy students have the opportunity to join research institutes for a few days.
In view of the health situation, everyone had to adapt. Nevertheless, the students were able to get to know, virtually, one research team per day, including the Inria Stars team, and thus participate in the project of Susanne Thummler and the team.

From the study of videos to algorithms

The quality of the algorithms depends on the accuracy of the video study.

The students were thus able to discover this work during their research internship and some will integrate the project to carry out their Master 2 thesis. And, most importantly, their first annotations made during this day will allow the Stars team to make good progress in the implementation of the first algorithms. It is therefore a well-defined process that has been put in place. The time required to study a video, in order to annotate it correctly and precisely, is at least a factor of 10 in relation to its duration. Each video will be analysed several times with a focus on a specific objective, in order to obtain the most complete possible annotation of the evaluation.

Two exemples:

The gaze: an autistic person has the peculiarity of not looking the Other in the face. It will then be necessary, for example, to identify what his gaze is like, if he looks for a long time, often, etc.
Gestures: it allows, among other things, to identify the repeated and stereotyped gestures of a child with autism, for example "flapping" (flapping of the arms and hands) or rocking of the body; whether these movements are present, frequent, prolonged, etc.

The precision of the video annotations provided by clinicians and specialists could then allow the creation of high-performance algorithms and an easier diagnosis of autism. Indeed, reducing the time spent by the clinician in analysing what can be automated would allow him to focus on other aspects and details of the patient's assessment.

They can also help with diagnosis:

for points that cannot be taken into account at the moment because of the large amount of existing and sometimes imperceptible information at first sight
by a quantification of abnormal movements such as stereotypes with an association with the severity of autism: mild, severe, etc.

in view of the amount of existing information in a video. A parallel could be drawn with the elements provided by an X-ray, a scanner, or an MRI.