Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects social communication and behaviour or how people communicate and interact with others. Diagnosing ASD typically relies on expert evaluations, which can be complex, time-consuming, and often delay early intervention. Early detection is crucial for better outcomes, as it provides opportunities to use brain plasticity during critical developmental periods.
Although there are no reliable biological markers for ASD yet, research has identified motor behaviour, such as atypical walking patterns (gait), as a potential early indicator. Research shows that people with ASD often have a unique way of walking characterized by shorter step lengths, reduced arm swings, and differences in shoulder movement. Traditional methods for assessing gait involve capturing the gait with a camera and sensors and then assessing for differences. These methods can be costly and intrusive, necessitating a more economical and less invasive approach.
Now, a study by UPES, Uttarakhand, and the Indian Institute of Technology (IIT) Kanpur explores the potential of using computer vision for gait analysis as an early detection tool for ASD.
Gait is considered an interplay of neurocognitive, locomotor, and learning factors with important brain systems involved in the activity. To analyse the subtle differences between the gait of two individuals, the researchers leveraged a technique called pose estimation, a type of machine learning algorithm, to analyze movements captured in videos.
This method allowed the researchers to use a regular video camera instead of expensive equipment and complex setups to capture the gait accurately. The data from the camera is analysed using software called MediaPipe, an open-source framework for analyzing and processing visual and audio data effectively. MediaPipe’s pose estimation tools help capture and analyze the gait by identifying and tracking key points on the individuals in video footage.
The study involved 32 children with ASD and 29 typically developing children. By capturing their gait on camera, the researchers measured their step length, and shoulder, and elbow movements. Using these measurements, they employed machine learning algorithms to identify patterns that could accurately distinguish between children with ASD and those without. Among the different machine learning models tested, the binomial logistic regression model was the most accurate, achieving an 82% success rate in classification.
Traditional gait analysis methods are often prohibitively expensive and sometimes require sensors to be attached to the body. This new method uses regular video cameras, making it more affordable and completely non-intrusive as it only requires video footage.
Despite its promise, this study has some limitations. It only involved a small number of participants, and larger studies with larger and more diverse sample sizes are needed to confirm these findings across a more diverse group of children. The researchers also relied on 2D video capture, which might miss subtle nuances in gait that could be caught with 3D analysis. Future work could explore enhancing current methods to better capture all aspects of gait. Further exploration is also needed into how gait links with other characteristics of autism to develop a comprehensive diagnosis plan.
This research represents a step forward in the early detection of ASD. By simplifying the process of gait analysis using innovative technology, a cost-effective and non-intrusive method for identifying children at risk of ASD can be offered. While further studies are necessary to overcome current limitations and broaden the research, this approach holds significant promise for transforming how ASD is detected and treated in the future.
This research news was partly generated using artificial intelligence and edited by an editor at Research Matters.