Algorithm shows high accuracy for detecting stereotypical autism movements in children

News
Article

A new study suggests an early AI-powered algorithm could help accurately identify common movements exhibited by children with autism spectrum disorder.

Algorithm shows high accuracy for detecting stereotypical autism movements in children

A deep learning-based algorithm designed to automatically identify and quantify stereotypical motor movements (SMMs) in children with autism spectrum disorder (ASD) was associated with a 92.5% sensitivity, according to a new analysis.1

The findings convey a potential breakthrough in objectively defining SMMs in children with ASD—a capability that may help deliver more large-scale research on and better clinical interpretation of the common physical trait of autism.

An international team of investigators conducted a retrospective cohort study in order to assess a new open-source artificial intelligence (AI) algorithm designed to analyze and define heterogenous SMMs in children with ASD. SMMs, the team noted, are an apparent physical trait in approximately half of all individuals with autism—presenting often as repetitive acts like flapping hands, rocking one’s body back and forth, and turning in circles.2

"SMMs are often described by individuals with ASD as an adaptive self-regulating coping mechanism for situations involving sensory overload, anxiety, or excitement,” investigators wrote. “However, frequent SMMs may also disrupt learning, skill acquisition, and social communication. Regardless of their function, identifying and quantifying SMMs is important for assessing SMM severity at diagnosis, estimating changes their severity over time, performing comparisons across developmental disorders, and for studying their underlying neurophysiology.”1

The algorithm used in the assessment was fed video recordings of 241 children with ASD, which included 580 hours’ worth of 319 behavioral assessments, including cognitive and language evaluations. A team of manual annotators identified 7352 video segments containing heterogeneous SMMs performed by children—comprising 21.4 hours of the total video.

Among the observed 241 children, 172 (78%) were male, and the mean age was 3.97 years old. The algorithm accurately detected 92.53% of all the manually annotated SMMs in the test data (95% CI, 81.09 – 95.10). However, it reported a specificity of 66.82%—showing a tendency for false positives (95% CI, 55.28 – 72.05).

The team observed that the overall number and duration of algorithm-identified SMMs per child significantly correlated with the manually annotated number and duration of SMMs.

“These correlations demonstrate the value of our algorithm for quantifying SMM severity per child,” the team wrote. “We believe our algorithm has the potential to replace manual annotation techniques previously applied to short recordings in small samples, thereby enabling large scale studies on a variety of topics, such as characterizing the development of SMMs in children with ASD and identifying their behavioral, physiological, and neural triggers.”

The trial findings were limited by the short age range of patients with ASD (1.4 – 8.0 years old), as well as a dearth of female-representative participants. Additionally, the currently used algorithm was not capable of distinguishing between different types of SMMs, and there was no assessment of the algorithm’s definition for SMM intensity onset and offset time accuracy.

All the same, the team concluded with high anticipation for what future iterations of this AI-powered algorithm could further provide to the clinical analysis of pediatric ASD.

“Our algorithm and ASDPose dataset offer an innovative way of studying SMMs in ASD and other disorders where individuals exhibit SMMs,” investigators wrote. “This novel digital phenotyping technique offers opportunities for studying the natural history of SMMs in autism as well as their underlying neural and physiological mechanisms.”

References

  1. Barami T, Manelis-Baram L, Kaiser H, Ilan M, et al. Automated Analysis of Stereotypical Movements in Videos of Children With Autism Spectrum Disorder. JAMA Network Open. Published September 12, 2024. 2024;7(9):e2432851. doi:10.1001/jamanetworkopen.2024.32851
  2. Melo C, Ruano L, Jorge J, et al. Prevalence and determinants of motor stereotypies in autism spectrum disorder: a systematic review and meta-analysis. Autism. 2020;24(3):569-590. doi:10.1177/1362361319869118
Recent Videos
Willough Jenkins, MD
Willough Jenkins, MD
Screening for and treating the metatarsus adductus foot deformity |  Image Credit: UNFO md ltd
Perry Roy, MD
Perry Roy, MD | Image Credit: Carolina Attention Specialists
Joanne M. Howard, MSN, MA, RN, CPNP-PC, PMHS & Anne Craig, MSN, RN, CPNP-PC
Donna Hallas, PhD, CPNP, PPCNP-BC, PMHS, FAANP, FAAN
Related Content
© 2024 MJH Life Sciences

All rights reserved.