ID 2332: Finding Impactful Movement Patterns for Automated GMA-scoring of Infants

Master’s Thesis

The quality of spontaneous movement in infants can give valuable insights into the health of their central nervous system. Hereby, the complexity and variety of such movements are crucial indicators and typically assessed by experts with the help of videos in the so called General Movement Assessment (GMA). Since the number of trained experts is limited, an automated approach could help spreading the application more widely.

Details

The goal of this thesis is to extract representative infant movement patterns and investigate their impact on GMA-scoring based on pose data from depth-cameras. This process should be entirely data-driven and without manual interaction. Hereby, varying length and appearance of the same movement type in different subjects are particularly challenging and need to be addressed. The extracted movement patterns should then be used to regress the GMA-score of previously unseen data. Additionally, the extracted movement patterns can be further investigated for a more objective score assessment.

Tasks

• Literature research
• Extract representative movement patterns across multiple subjects
• Investigate the impact of the individual types of movement on GMA-scoring
• Create a dictionary of representative movements with positive or negativ impact
• Design a system for regressing the GMA-score based on this dictionary of movements

Requirements

• Strong programming skills in Python
• Advanced knowledge and some experience in Machine Learning
• Advanced knowledge in signal processing
• Bachelor’s degree in a related field of study

Supervisors

Philip Stoll, M. Sc.

Researcher & PhD Candidate

Matthias Zürl, M. Sc.

Researcher & PhD Candidate