Machine learning for video-based assessment of movement disorders: application to early onset ataxia and developmental coordination disorder
Machine learning for video-based assessment of movement disorders
Children with movement disorders often struggle with balance, coordination, and everyday motor tasks. Two common conditions Early Onset Ataxia (EOA) and Developmental Coordination Disorder (DCD) can look very similar in the clinic, making it hard for doctors to tell them apart. A precise diagnosis is crucial, since early treatment can improve a child’s long-term development.
Traditionally, doctors rely on visual observation and rating scales to assess these disorders, but such methods can be subjective and vary between clinicians. This thesis of Wei Tang explores how artificial intelligence (AI) and video analysis can help make these assessments more objective and consistent. Using short videos of children performing simple tasks such as walking or touching their finger to their nose, the research develops AI systems that automatically recognize patterns of movement linked to EOA, DCD, or healthy motor function.
The thesis combines traditional machine learning and modern deep learning models to analyze movement from regular clinical videos, without the need for special sensors or equipment. These models achieved promising accuracy and were able to highlight which parts of a child’s movement were most important for their decisions, making the results more interpretable for clinicians.
By focusing on four key qualities, reproducibility, accuracy, explainability, and robustness, this work moves a step closer to trustworthy, AI-based tools that can support doctors in diagnosing pediatric movement disorders. In the future, such systems could make neurological assessments faster, more objective, and more accessible for children and families worldwide.