Automated Quantification of Stereotypical Motor Movements in Autism Using Persistent Homology.

Area of Science:

• Neuroscience
• Data Science
• Biomedical Engineering

Background:

• Stereotypical motor movements (SMM) are key indicators in autism diagnosis.
• Current methods for quantifying SMM lack efficiency and broad applicability across individuals and developmental stages.

Purpose of the Study:

• To develop a novel pipeline for quantifying and characterizing recurrent movement patterns in autism using Topological Data Analysis (TDA).
• To create interpretable feature vectors that capture geometric properties of autistic SMM from time-series data.

Main Methods:

• Utilized persistent homology from TDA to analyze time-series data from video pose estimation and wearable sensors.
• Extracted periodic structures to generate low-dimensional feature vectors representing movement patterns.
• Employed simple classifiers for automated SMM quantification.

Main Results:

• The TDA-based features enabled accurate automated quantification of autistic SMM.
• Visualizations showed that extracted features are generalizable across individuals and not person-specific.
• The pipeline demonstrated potential for scalable and interpretable characterization of SMM.

Conclusions:

• Topological Data Analysis offers a mathematically principled approach to characterizing autistic SMM.
• This method supports person-agnostic and scalable quantification of SMM in naturalistic settings.
• The developed pipeline shows promise for advancing autism research and diagnostics.