A Muscle Pennation Angle Estimation Framework From Raw Ultrasound Data for Wearable Biomedical Instrumentation
Wearable ultrasound pennation analysis
Abstract
Muscle pennation angles are key biomarkers for musculoskeletal research and rehabilitation. This work introduces a framework that estimates pennation angles directly from raw 32-channel ultrasound data using feature extraction and an XGBoost regressor aligned with automatic annotations. The method delivers near real-time predictions with a root-mean-square error of 1.6° while compressing the model footprint to 11 kB, enabling execution on a GAP9 microcontroller with 1.31 ms latency and 1.03 mJ energy consumption. The approach unlocks wearable biomedical instrumentation for continuous muscle analysis.
Key Highlights
- Estimates muscle pennation angles directly from raw 32-channel ultrasound using an XGBoost regressor with automatic annotations.
- Achieves ~1.6° RMSE with an 11 kB model, 1.31 ms inference latency, and 1.03 mJ energy per prediction on GAP9.
- Enables wearable biomedical instrumentation for continuous muscle analysis and rehabilitation monitoring.
Thorir Mar Ingolfsson
Postdoctoral Researcher
I develop efficient machine learning systems for biomedical wearables that operate under extreme resource constraints. My work bridges foundation models, neural architecture design, and edge deployment to enable real-time biosignal analysis on microwatt-scale devices.