TL;DR
Researchers have validated a 1-minute, videogame-like tool called the Computerised Assessment of Motor Imitation (CAMI) as a specific biomarker for Autism Spectrum Disorder (ASD). Unlike traditional diagnostic tools that can be "clouded" by co-occurring conditions, CAMI successfully distinguishes ASD from both neurotypical development and ADHD by identifying unique motor imitation deficits specific to the autism phenotype.
What Problem Does This Paper Address?
Diagnosing ASD is often a complex, lengthy process, frequently delayed by the presence of co-occurring conditions like Attention-Deficit Hyperactivity Disorder (ADHD), which affects 50–70% of children with autism. While motor imitation—the ability to copy someone else's physical movements—has long been recognized as a potential biomarker for ASD, traditional assessments rely on subjective, time-consuming human observation.
The authors address two major gaps: the lack of a scalable, objective tool for measuring imitation and the need to determine if imitation deficits are truly "autism-specific" or simply a general feature of neurodevelopmental delay shared with conditions like ADHD.
Methods
The study introduces CAMI, a brief (1-minute) task where children are asked to copy the dance-like, whole-body movements of a video avatar.
- Hardware: The system uses two sensorless Kinect Xbox cameras to record movements from the front and back.
- Algorithm: Computer vision methods automatically identify key joints and calculate an imitation score (0 to 1) by comparing the child's movements to the avatar's in a data-driven manner.
- Validation: The researchers used four levels of analysis:
- Linear Regression: To examine group differences while controlling for IQ and gender.
- Dimensional Analysis: Correlating CAMI scores with established clinical measures of autism, ADHD, and motor skills.
- ROC Curves: To test the diagnostic sensitivity and specificity.
- Support Vector Machines (SVM): To see if a simple machine learning model could classify children based solely on their CAMI scores.
Datasets
The study involved a sample of 183 children (aged 7–13) categorized into four groups:
- ADHD without ASD (n=35)
- ASD with co-occurring ADHD (n=63)
- ASD without ADHD (n=20)
- Neurotypical (NT) development (n=65)
Performance was benchmarked against "gold-standard" clinical batteries, including the ADOS-2 (Autism Diagnostic Observation Schedule), SRS-2 (Social Responsiveness Scale), Conners rating scales for ADHD, and the PANESS for physical motor coordination.
Key Insights and Ideas
- The "Specificity" Discovery: Children with ASD showed significantly poorer imitation performance regardless of whether they also had ADHD. Crucially, children with ADHD performed similarly to neurotypical children, suggesting that imitation is a highly specific marker for the autism phenotype rather than a general marker of neurodivergence.
- Divergent Pathways: In neurotypical children, better imitation was linked to better attention and general motor ability. However, in children with ASD, imitation performance was specifically tied to social affect and repetitive behaviors. This suggests that for autistic children, the struggle to imitate is rooted in social-communicative brain circuits rather than just "clumsiness" or lack of focus.
- Scalability vs. Subjectivity: By removing the human coder from the loop, CAMI provides a high-precision, objective metric that can be deployed in minutes, potentially serving as a "vital sign" for neurodevelopment in clinical settings.
Key Results
- Diagnostic Accuracy: CAMI distinguished ASD from neurotypical children with an 80% true positive rate (AUC = 0.80) and from children with ADHD with a 70% true positive rate (AUC = 0.70).
- SVM Classification: The machine learning model achieved 71.6% accuracy in identifying ASD vs. NT and 66.4% accuracy for ASD vs. ADHD using only the 1-minute imitation data.
- Statistical Significance: The difference in imitation scores between ASD groups and the NT group was highly significant ($P < 0.0001$), with a large effect size (Cohen’s $d$ between 1.13 and 1.20).
Limitations and Future Work
The authors acknowledge several areas for further development:
- Sample Characteristics: The current study focused on "high-functioning" children (IQ > 70). Future iterations need to simplify instructions to accommodate children with intellectual disabilities.
- Sample Size: The "ASD-only" group was relatively small (n=20), which may have limited the power to detect subtle differences between ASD subgroups.
- Age Range: While effective for 7–13 year-olds, the tool needs to be validated in younger toddlers, where early intervention is most critical.
- Longitudinal Data: Future research should investigate if improving CAMI scores through intervention correlates with improvements in real-world social communication skills.
Summary updated