Analysis of facial expressions and voice in children: a contact-less support to the diagnosis of autism spectrum disorders

Abstract

Autism Spectrum Disorders (ASD) are neurodevelopmental disorders that impair the social- communicative ability, motor and behavioural skills from very early childhood. The aetiology is not defined yet, but both genetic and environmental components are presumed. Prevalence values are constantly increasing and vary between 1/68 in the USA and 1/150 in Europe. A correct and timely diagnosis, usually made between 18 and 48 months of life, is of basic relevance for an adequate rehabilitative care. Therefore, early identification of ASD represents an important challenge in the clinical field.

To date the diagnosis is based on a qualitative assessment during short sessions carried out in a small room and lasting 20’-30’. The child is "observed" in his/her emotional reactions to specific stimuli, in the interaction with the adult and the surrounding environment through the Autism Diagnostic Observation Schedule (ADOS-2) protocol, considered as the gold standard method to screen children with suspected ASD. During the ADOS-2 test, the observer, usually a psychologist, looks for bizarre behaviours in the communicative field, social interaction, flexibility/rigidity, and gives scores according to the ADOS-2 classification scale.

Communication deficits are evidenced by the lack or reduced verbal and non-verbal communication. Among the symptoms of ASD, subjects show abnormal or diminished use of facial expressions and anomalies in verbal communication that, in the most serious cases, occurs only through the emission of few vowel sounds.
Therefore, suitable techniques based on automated contact-less systems for video and audio analysis can provide an objective aid to the diagnosis in such critical subjects, avoiding conditioning of the child and therefore possible bias of the diagnostic investigation.

This is the main subject of a study under development within the POR-FESR RSI 2014-20 project “System for supporting clinical practice for the assessment and monitoring of Autism Spectrum Disorders in the diagnostic phase (DI-ASD)”, funded by the Tuscany Region. DI-ASD is a project carried on at the Biomedical Engineering Laboratory along with the Stella Maris Foundation Hospital (Pisa, Italy) and three small companies based in Tuscany.

Once completed, this innovative system could support clinicians during the ADOS-2 test providing objective data and statistics. It would also provide a database of labelled facial expressions of 2-6 years old children, both typically developed and ASD, that to date does not exist. Such database could help clinicians in detecting atypical patterns setting up personalized therapeutic treatments and follow-up not only in the ASD, but also in other neurodevelopmental disorders.

Specifically, the ongoing research at the Biomedical Engineering Laboratory concerns two main aspects:
1. Analysis of the four basic facial expressions (happiness, surprise, sadness, neutral) that are directly related to the emotional state of the subject.
2. Analysis of main acoustical parameters and prosody in: pre-verbalizations, vocalic emissions and speech of pre-school and school children.
Implemented methodologies and preliminary results will be shortly presented.