[ad_1]
According to a new study from the Faculty of Medicine at Stanford University, short home videos can be used to diagnose autism in children.
The research, which will be published online on November 27 in PLOS Medicine, develops a feasibility study conducted in 2014 on the subject by the same researchers. In the new study, scientists used machine learning to determine the characteristics of children's behavior to badess to badess autism, using a computer to reduce a long list of behavioral characteristics to those most relevant to the diagnosis. They also designed an algorithm that weights each characteristic to provide an overall diagnostic score for each child.
"In the United States, the average waiting list for access to standard care can last up to one year," said lead author of the study, Dennis Wall, PhD, badociate professor of pediatrics and biomedical data science at Stanford. "The use of personal videos for diagnostic purposes can potentially streamline the process and make it much more efficient."
Personal videos offer another potential benefit for the diagnosis of behavioral and developmental disorders such as autism. "Home video captures the child in his natural environment," Wall said. "The clinical environment can be brutal and artificial, and can lead to atypical behavior on the part of children."
Value of early diagnosis
Autism is a developmental disorder characterized by limited interests, repetitive behaviors and difficulties in creating social bonds. Previous research has shown that behavioral therapies for autism work best when they are started before the age of 5, but long waiting lists for testing make it difficult for children to have autism. 39, family access to treatment in a timely manner. Current diagnoses take a lot of time and require an individual badessment with an autism specialist. Clinicians spend a few hours per patient evaluating dozens of aspects of the child's behavior.
In the new study, researchers designed and tested eight models of machine learning for the diagnosis of autism from short videos. Each model consisted of a set of algorithms comprising five to twelve characteristics of the children's behavior and produced an overall numerical score indicating whether the child was autistic.
To test the models, the researchers asked families recruited via social networks and list servers for autism to submit brief videos at home. 116 videos of autistic children (average age of 4 years and 10 months) and 46 videos of typical developing children (average of 2 years old) were also received. years, 11 months) that met their criteria: The videos lasted 1 to 5 minutes and showed the child's face and hands, showed a direct social engagement or opportunities to engage and showed opportunities for use objects such as toys, colored pencils or utensils.
Nine video reviewers received a brief instruction on how to evaluate each video, answering 30 questions yes / no about whether the children in these videos exhibited certain behaviors such as the use of 39, expressive language, eye contact, the expression of emotion and the fact of attracting attention to objects. All yes / no questions were based on the behavioral characteristics used in standard autism screening tools.
The nine evaluators scored 50 videos and the researchers used these results to determine that three evaluators were the minimum number required to generate a reliable score. The remaining videos were randomly badigned to the reviewers, with three reviewers badigning a score to each video.
On average, watching and rating the videos took 4 minutes each. The data from each video, consisting of 30 yes / no answers to questions about the child's behavior, were incorporated into the eight mathematical models.
One model, a logistic regression model that used five behavioral characteristics, gave the best results, identifying autism with an overall accuracy of 88.9%, including correct labeling of 94.5% of autistic children and children with autism. 77.4% of children without autism.
To validate their findings, the researchers repeated the experiment with 66 additional videos: 33 with children with autism and 33 with non autistic children. The same model worked even better, with a correct identification of 87.8% of autistic children and 72.7% of non autistic children.
"We showed that we could identify a small set of behavioral characteristics aligned with the clinical outcome, that non-experts could quickly and independently badign a score to these features in an online virtual environment in minutes, and that the model used to combine these features was: effective to produce a score corresponding to the clinical outcome, "Wall said. The final scores are not just a diagnosis "yes or no" of autism, he added; Instead, numerical scores can contain information about the severity of the disorder and be useful for tracking progress over time.
Provide a tool for pediatricians
Wall hopes that simple scoring systems for personal videos will help streamline the autism diagnostic process. "This could be used in general pediatric settings such as baby checkups," he said, adding that video scores could be plotted over time and compared to those of the general population, the same way that the height and weight of the child are represented on a growth chart. .
"Our long-term dream is that a tool like this gives pediatricians generally more confidence to make diagnostic decisions regarding autism and other developmental disorders," he said. -he declares. For a very young child – at an age when autism may be hard to distinguish from normal development – the doctor's decision may be to wait under supervision, but with the benefit of having access to treatment. a video score as a baseline for subsequent evaluations. In other cases, it can be clear that a child should start autism treatment immediately or should be referred to a specialist for more detailed diagnostic evaluation.
The researchers are now repeating their investigation with home-grown videos of young children in Bangladesh to see how their mathematical models translate across cultures.
###
The main author of the study is Qandeel Tariq, data badyst in the Wall Lab. Other Stanford authors are Jena Daniels, a former director of clinical operations; Jessey Nicole Schwartz, Clinical Research Coordinator; Peter Washington, graduate student in bioengineering; and postdoctoral researcher Haik Kalantarian, PhD. Wall is a member of Stanford Bio-X, the Stanford Child Health Research Institute and the Wu Tsai Neuroscience Institute.
The research was funded by the National Institutes of Health (grants 1R01EB025025 and 1R21HD091500), the Hartwell Foundation, the Bill and Melinda Gates Foundation, the Coulter Foundation, the Lucile Packard Children's Health Foundation, and the Integrated Diagnostics Program. Stanford Center, Beckman Center, Bio-X, the Accelerator Forecast and Diagnostics Program, and the Institute for Research on Child Health. The research also received support from David Orr, Imma Calvo, Bobby Dekesyer and Peter Sullivan.
The pediatric and biomedical science departments at Stanford also supported the work.
The Faculty of Medicine at Stanford University consistently ranks among the nation's top medical schools, integrating research, medical education, patient care and community services. For more information on this school, go to http: // med.
Press contact: Erin Digitale at (650) 724-9175 ([email protected])
Contact for Audiovisual Media: Margarita Gallardo at (650) 723-7897 ([email protected])
Source link
