Influence of the affinity on people with autism spectrum disorders – Computational models of visual perception

Context

The exploratory action, funded by CominLabs for a 6-month period from March to October 2019, has been a unique opportunity to test our main hypothesis: does a digital affinity displayed on screen influence the visual engagement of the ASD subject?

This action aims to the definition of an innovative eye tracking protocol. The set of displayed images was composed of 25 neutral images and 5 affinity images specific to each participant. Once the preparation was over, the eye tracking experiment could be performed. This was done by a computer scientist, in presence of the 2 psychologists and the MIE (Medical Educational Institutes) staff.

Although many hurdles have been overcome during the exploratory action, we have to go further by revisiting in-depth the protocol we have proposed. Many aspects should be thoroughly checked and would very likely required adaptations or further developments. For instance, we do not measure and compare results with a control group due to the lack of time. We do not evaluate neither the situational anxiety nor the level of stress. In addition, we do not investigate the benefit to make the tests several times for people with autism. Finally, it would also be interested to evaluate whether or not the presence of an adult well known by the subject, such as caregivers, generates a reaction.

Objectives

For going further, we aim to investigate the following points:

• Multimodal analysis
  The use of an eye tracking apparatus presents some limitations, especially when the subject disengages his attention. In addition, the capability to track the gaze can be difficult in some cases. To deal with this limitation, a new protocol would consist in recording data from different modalities such as the gaze, the voice and the activity of the subject
• Mining information and computational models of saliency for autisms
  Thanks to collected data, we aim to redefine saliency model, thanks to convolutional neural network and deep learning [Nebout et al., 2019; Wei et al., 2019], for quantifying the saliency difference between people with autism and a matched group control. It would indicate the relative contribution of image attribute to the gaze deployment.
• Panel of subjects
  We also want to consider several kinds of autism spanning from Kanner to Asperger’s autism. For all kinds of clinical orientations, this large variety of autism is commonly referred to as continuum of autism (this does not mean there is a “genius” behind each subject with autism).
• Towards an ecological experiment
  To generalize our findings, we would like to define an experiment by using eye-tracking glasses. This kind of eye tracker allows us to be closer to the real world, and then to be more ecological, reducing biases due to a controlled environment and to laboratory settings. Obviously, the use of eye tracking glasses requires to revisit in-depth the protocol on a case-by-case basis.