American researchers asked a group of families to shoot their children interacting with objects and people. They tried eight models of automated learning to diagnose autism, which allows “to optimize the process and make it much more efficient”, according to a study published in the scientific journal PLOS Medicine.
The study was developed by a team from Stanford University School of Medicine and was headed by a professor of pediatrics and biomedical science, Denis Wall, in this California city.
Each model contained a “set of algorithms that included from 5 to 12 behavioral characteristics of children and he compiled a total score that indicated whether the child had autism, ”he explained.
How the videos were processed
Wall said that in order to evaluate the models, they asked the families recruited for the study to send home videos of between one and five minutes. which showed children's faces and hands, as well as their “social interaction, as well as the use of toys, pencils and dishes”., Of these images, 116 boys with an average age of 4 years and 10 months were diagnosed with autism, and another 46 (on average two years and 11 months) developed it, he explained.
Nine expert experts analyzed the video using 30 questionnaire with “yes” or “no” answers based on the typical behavior of autism, which were then included in eight mathematical models.
The model that showed the best results is that which revealed 94.5% of cases of children with autism and 77.4% of children without autism. To test the results they rated 66 other videos, half of them are children with autism. In the same model, 87.8% of cases of children with autism and 72.7% of those who did not have this were correctly identified.
Another advantage of using home movies for diagnostics is that “they take the child in their natural environment,” in contrast to the clinical evaluation conducted in an environment “that can be tough and artificial and cause atypical behavior.” “We have shown that we can identify a small group of behavioral characteristics that are very close to clinical outcomes, and that non-specialists can quickly and independently evaluate these characteristics in a virtual online environment in minutes,” said Wall.