The US researchers asked a group & # 39; families to launch their children by interacting with & # 39; goods and people. they tried eight models & # 39; automated learning for the diagnosis & # 39; autism, enabling "streamlining the process and making it more effective", according to the study published in the scientific journal PLOS Medicine.
The study was developed by a team from the University of Medicine School & # 39; Stanford and was led by Dennis Wall, professor of Paediatrics and Information Science of Biomedical & # 39; that the Californian city.
Each of the models contained "set & # 39; Algorithms included 5 to 12 feature & # 39; child behavior and produced an overall score indicating whether children have autism ", explained.
How dealt with videos
Wall said that to evaluate the models, asked the families recruited for the study to send home videos & # 39; between forty five minutes. showcasing the faces and hands of children caught "the social interaction and the use of & # 39; toys, pencils and tools".. From these images 116 children with & # 39; average age & # 39; 4 years and 10 months were diagnosed with autism and 46 other (b & # 39; average & # 39; years and 11 months) develop it were, he explained.
Nine & # 39; expert reviewers analyzed using videos & # 39; 30 questions questionnaire with answers "yes" or "no", based on typical behaviors of & # 39; autism, which were then incorporated into the eighth mathematical models.
The model offered the best results is identified that 94.5% of cases & # 39; & # 39 with children, autism and 77.4% of & # 39; those & # 39; children without autism. To verify the results evaluated 66 other videos, Half of & # 39; children with autism. The same pattern identified by & # 39; correct manner 87.8% of cases & # 39; & # 39 with children, autism and 72.7% of & # 39; those without this disorder.
Another advantage of & # 39; & # 39 use; of – for home video – diagnosis is to "take children in their natural environment", b & # 39; difference from the clinical assessment carried out in & # 39; medium "can & # 39; be rigid and artificial and cause atypical behavior". "We showed that we can identify a small group & # 39; features & # 39; behaviors that are aligned well with & # 39; clinical results and non-experts to assess these features quickly and with & # 39; independently f & # 39; online virtual environment, f & # 39; minutes ", said Wall.