This research transforms how we analyze daily movements of Dementia patients through a novel two-stage self-supervised learning approach that converts activity data into meaningful behavioral insights.

• Converts time-series activity data into text strings for advanced pattern recognition
• Leverages fine-tuned language models to encode patient movement behaviors
• Uncovers deeper patterns in patient activity with high temporal precision
• Enables more personalized care interventions based on detected behavioral changes

Medical Impact: By transforming raw activity data into interpretable patterns, healthcare providers can detect subtle changes in patient behavior earlier, potentially allowing for more timely and targeted interventions for Dementia patients living at home.

Analyzing Patient Daily Movement Behavior Dynamics Using Two-Stage Encoding Model