Autism, artificial intelligence helps
Artificial intelligence can help quantify autism spectrum disorders.
This is the conclusion of a joint study carried out by ISASI-CNR in Lecce and the Department of Human Studies of the University of Salento, in collaboration with the Institute of Communication and Health di Lugano that was published in Sensors.
Within the 'si-robotics' project, researchers conducted a long-time study in the Salento area and obtained interesting results thanks to modern artificial intelligence techniques, including artificial vision and deep learning.
By using low-cost technology such as webcams, which can analyze frames and measure a child’s facial movements, researchers have developed a method to evaluate the expressions of autistic children when they reproduce emotions, by measuring the contraction and distension of facial muscles through the localization and tracking of fiducial landmarks of the face.
In this way, it was possible to evaluate the micro-expressions, typical of autism, that usually emotion facial recognition software programmes do not detect since they are based on subjects with normal development.
Moreover, the researchers observed that human beings cannot detect expressive changes of the upper part of the face, if the subject observed has an expressiveness based on marked movement of the mouth.
This strong restriction of our visual system alters our evaluation, while machines can detect even micro-movements at the pixel level.