Adaptive P-Splines for challenging filtering problems in biomechanics.

Area of Science:

• Biomechanical analysis
• Signal processing
• Computational statistics

Background:

• Noise suppression is crucial for biomechanical data analysis.
• Existing methods like low-pass filters and smoothing splines struggle with signals exhibiting rapid derivative changes, common in impact events.
• These traditional methods often use fixed parameters that fail to adapt to dynamic signal characteristics.

Purpose of the Study:

• To introduce and evaluate adaptive penalized splines (APS) for processing biomechanical signals with rapidly changing derivatives.
• To compare the performance of APS against conventional filtering techniques, specifically the Butterworth filter.
• To assess the ability of APS to provide uncertainty estimates for derived quantities.

Main Methods:

• Examined three variations of APS: independent random variables, autoregressive process, and smooth cubic spline.
• Inferred temporal adaptations in regularization penalty from observed data.
• Compared APS performance against a 4th order Butterworth filter using simulated and real-world impact data.

Main Results:

• APS demonstrated significant Root Mean Square Error (RMSE) reductions of 48%-183% compared to Butterworth filtering on simulated data.
• For pendulum impact data, APS achieved 13%-28% RMSE reduction in acceleration inference compared to an optimally tuned Butterworth filter.
• APS provided reliable uncertainty estimates for fitted curves and derived metrics like peak accelerations.

Conclusions:

• Adaptive penalized splines (APS) are highly effective for analyzing biomechanical signals with rapid changes, outperforming traditional methods.
• APS offer improved accuracy in capturing transient events like impacts and provide valuable uncertainty quantification.
• Researchers analyzing impact events or requiring precise derivative estimation should consider employing APS for enhanced data analysis.