TL;DR
This paper introduces a novel two-stage algorithm for real-time estimation of pathological tremor parameters using gyroscope data. This advancement is crucial for developing advanced neuroprosthetics that can selectively compensate for tremor while preserving voluntary movement, ultimately improving the quality of life for individuals with movement disorders.
What Problem Does This Paper Address?
Pathological tremors, such as those seen in Parkinson's disease or essential tremor, significantly impair daily activities and social functioning. While current treatments like medication, surgery, and deep brain stimulation have limitations (side effects, limited efficacy), a promising alternative is functional compensation through biomechanical loading. This approach requires precise, real-time knowledge of tremor characteristics (amplitude and frequency) to effectively counteract tremor without hindering voluntary movements. Traditional tremor recording methods have limitations, and existing estimation algorithms often struggle with the low signal-to-noise ratio of tremor compared to voluntary motion. This paper aims to bridge this gap by developing a robust real-time system for estimating tremor parameters from gyroscope data.
Methods
The core of this work is a two-stage algorithm designed for real-time tremor parameter estimation:
- Stage 1: Voluntary Motion Tracking and Tremor Isolation: This stage leverages the fact that voluntary movements and pathological tremors have distinct frequency distributions. Gyroscope data, representing total motion, is first processed to estimate the voluntary component. This is achieved using a Critically Dampened (CDF) g-h filter, which was found to outperform other candidates (Benedict-Bordner Filter and Kalman Filter) in terms of tracking accuracy and smoothness. The estimated voluntary motion is then subtracted from the total motion to isolate the tremorous component.
- Stage 2: Tremor Parameter Estimation: The isolated tremor signal is then fed into two algorithms:
- Weighted Frequency Fourier Linear Combiner (WFLC): This algorithm estimates the instantaneous tremor frequency.
- Kalman Filter (KF): This algorithm uses the frequency information from the WFLC to estimate the instantaneous tremor amplitude.
This cascaded approach aims to provide robust and accurate real-time estimates of both tremor frequency and amplitude.
Datasets
The study utilized data collected from five patients diagnosed with common tremor-causing pathologies. These patients underwent four specific clinical and functional tasks designed to elicit different types of tremor (postural, kinetic, rest) and assess functional abilities. The data was recorded using MEMS gyroscopes at a sampling frequency of 1 kHz. No new benchmark datasets were introduced; rather, existing patient data was used for evaluation.
Key Insights and Ideas
- Frequency Domain Separation is Key: The fundamental idea that voluntary motion and pathological tremor occupy different frequency bands is the cornerstone of the algorithm, enabling their separation.
- Gyroscope Advantage: Gyroscopes offer a direct measurement of rotational speed, overcoming limitations of accelerometers which are influenced by gravity and linear acceleration.
- Critically Dampened Filter for Voluntary Motion: The CDF proved to be an effective and robust filter for isolating voluntary motion, outperforming other tested filters due to its oscillatory nature that resonates with tremor, leading to better phase alignment.
- Cascaded WFLC-KF for Tremor: Combining a WFLC for frequency tracking with a KF for amplitude estimation provides a powerful and accurate system for real-time tremor parameter estimation, especially for the more variable amplitude parameter.
- Tremor as Additive Noise: The assumption that pathological tremor acts as additive noise to voluntary motion simplifies the estimation problem and allows for effective subtraction of the voluntary component.
Key Results
- The proposed two-stage algorithm achieved an average tremor estimation error of 0.001 ± 0.002 rad/s for amplitude.
- Frequency estimation was found to be robust and agreed well with spectrogram analysis.
- The Critically Dampened Filter (CDF) demonstrated superior performance in voluntary motion tracking, exhibiting the least Kinematic Tracking Error (KTE) across various tasks.
- The cascaded WFLC-KF approach for tremor parameter estimation resulted in a Filtered Mean Square Error with Delay correction (FMSEd) that was at least five times smaller than that obtained with the Bandlimited Multiple Fourier Linear Combiner (BMFLC), the second-best candidate.
Limitations and Future Work
- Generalizability: While validated on patients with major pathologies, the algorithm's performance across a wider range of tremor severities and types, as well as its generalizability to other joints, would require further investigation.
- Fixed Gain Limitations: The paper notes that algorithms like WFLC, with fixed gains, struggle with rapid changes in tremor characteristics, highlighting the advantage of the KF's adaptive gain.
- Future Work: The authors explicitly state that the real-time tremor parameters derived from this algorithm will be employed to drive a "soft robot" or neuroprosthesis for tremor suppression using Functional Electrical Stimulation (FES). This suggests that the immediate future work involves integrating this estimation system into a closed-loop control system for tremor compensation.
Summary updated