Gait parameter fitting and adaptive enhancement based on cerebral blood oxygen information

Haozhe Ma, Chunguang Li, Yufei Zhu, Yaoxing Peng, Lining SunFrontiers in Human Neuroscience 2023 · DOI

Abstract

Accurate recognition of patients' movement intentions and real-time adjustments are crucial in rehabilitation exoskeleton robots. However, some patients are unable to utilize electromyography (EMG) signals for this purpose due to poor or missing signals in their lower limbs. In order to address this issue, we propose a novel method that fits gait parameters using cerebral blood oxygen signals. Two types of walking experiments were conducted to collect brain blood oxygen signals and gait parameters from volunteers. Time domain, frequency domain, and spatial domain features were extracted from brain hemoglobin. The AutoEncoder-Decoder method is used for feature dimension reduction. A regression model based on the long short-term memory (LSTM) model was established to fit the gait parameters and perform incremental learning for new individual data. Cross-validation was performed on the model to enhance individual adaptivity and reduce the need for individual pre-training. The coefficient of determination (R2) for the gait parameter fit was 71.544%, with a mean square error (RMSE) of less than 3.321%. Following adaptive enhancement, the coefficient of R2 increased by 6.985%, while the RMSE decreased by 0.303%. These preliminary results indicate the feasibility of fitting gait parameters using cerebral blood oxygen information. Our research offers a new perspective on assisted locomotion control for patients who lack effective myoelectricity, thereby expanding the clinical application of rehabilitation exoskeleton robots.

Introduction

Lower-limb exoskeletons are transformative for rehabilitation, but they require a way to "read" a patient's movement intentions. Traditionally, this is done using electromyography (EMG) to detect electrical activity in the muscles. However, many patients with severe motor dysfunction—such as those with advanced paralysis or muscle atrophy—exhibit "silent" or highly corrupted EMG signals.

While researchers have looked toward brain signals (EEG or fNIRS) as an alternative, most existing studies only classify discrete states (e.g., "resting" vs. "walking"). To truly assist a patient in a natural way, an exoskeleton needs to adjust to continuous parameters: How fast does the person want to go? How long should the next stride be? This paper tackles the challenge of fitting these continuous variables using the relatively slow-moving signal of cerebral hemodynamics.

Methodology

The authors designed a pipeline that transforms raw brain oxygenation data into actionable gait commands:

  1. Data Acquisition: They used functional near-infrared spectroscopy (fNIRS) to monitor oxygenated (HbO), deoxygenated (HbR), and total hemoglobin (HbT) in the prefrontal cortex (PFC), supplementary motor area (SMA), and primary motor cortex (PMC). Simultaneously, inertial sensors (Xsens) captured ground-truth gait data.
Layout of the headgear and probe placement used for collecting cerebral blood oxygen signals.
  1. Feature Engineering: A massive set of 1,035 features across time, frequency, and spatial domains was extracted.
  2. Feature Selection & Compression: To handle this high-dimensional data, they used a two-step process: filtering based on Pearson correlation and Gradient Boosting Tree importance scores, followed by an AutoEncoder-Decoder neural network to compress the remaining features.
  3. The Model: A Long Short-Term Memory (LSTM) regression model was chosen for its ability to handle time-series data and capture the temporal dependencies between brain activity and physical movement.
  4. Adaptive Enhancement: Recognizing individual brain variability, they implemented incremental learning. This allows the model to start with a general "base" and then fine-tune itself using a small amount of data from a new individual, significantly boosting accuracy without requiring extensive pre-training.

Experimental Design

The study recruited 38 healthy volunteers (24 males, 14 females) aged 19–22.

  • Training/Testing Set: 30 subjects were used for the initial model development (8:2 ratio).
  • Adaptive Pool: 8 subjects were reserved specifically to test the adaptive enhancement strategy.
  • Experimental Tasks: Participants performed walking trials with adjusted gait velocity and adjusted stride length to provide a diverse range of continuous motion data.

Key Results and Discussion

The model demonstrated strong performance in fitting continuous gait parameters:

  • Gait Speed: Achieved a coefficient of determination ($R^2$) of 71.55% and a Root Mean Square Error (RMSE) of 1.36%.
  • Step Length: Achieved an $R^2$ of 72.02% and an RMSE of 3.32%.
  • Adaptive Boost: After applying the adaptive enhancement strategy, the $R^2$ for individual step parameters increased by an average of 6.985%, while the RMSE decreased by 0.303% (speed) and 1.198% (step length).

Insights

  • Hemodynamics as a Proxy for Intent: The study proves that even though blood oxygen changes are slower than electrical spikes (EEG), they contain enough high-resolution information to predict physical gait parameters.
  • Frequency Band Importance: Different frequency bands correlate with different gait aspects. Neural activity and endothelial metabolic activity bands were most important for speed, while heart rate and respiratory bands were more influential for step length.
  • Personalization: The adaptive enhancement results suggest a future where a patient could "calibrate" their exoskeleton in just a few minutes using incremental learning.

Conclusion

While promising, the study has hurdles to clear before clinical deployment. The participants were all healthy young adults; brain signals and gait patterns change significantly with age and pathology, requiring testing on mobility-impaired populations. Additionally, because fNIRS signals have an inherent hemodynamic delay, future work must address real-time prediction to compensate for this lag. Currently, the model fits the maximum value of a gait cycle rather than instantaneous flow, an area for future optimization to reduce false positives within a single gait cycle.

Summary updated

Abstract

Accurate recognition of patients' movement intentions and real-time adjustments are crucial in rehabilitation exoskeleton robots. However, some patients are unable to utilize electromyography (EMG) signals for this purpose due to poor or missing signals in their lower limbs. In order to address this issue, we propose a novel method that fits gait parameters using cerebral blood oxygen signals. Two types of walking experiments were conducted to collect brain blood oxygen signals and gait parameters from volunteers. Time domain, frequency domain, and spatial domain features were extracted from brain hemoglobin. The AutoEncoder-Decoder method is used for feature dimension reduction. A regression model based on the long short-term memory (LSTM) model was established to fit the gait parameters and perform incremental learning for new individual data. Cross-validation was performed on the model to enhance individual adaptivity and reduce the need for individual pre-training. The coefficient of determination (R2) for the gait parameter fit was 71.544%, with a mean square error (RMSE) of less than 3.321%. Following adaptive enhancement, the coefficient of R2 increased by 6.985%, while the RMSE decreased by 0.303%. These preliminary results indicate the feasibility of fitting gait parameters using cerebral blood oxygen information. Our research offers a new perspective on assisted locomotion control for patients who lack effective myoelectricity, thereby expanding the clinical application of rehabilitation exoskeleton robots. This work establishes a foundation for promoting the application of Brain-Computer Interface (BCI) technology in the field of sports rehabilitation.
BibTeX
@inproceedings{ma2023gait,
  author = {Haozhe Ma and Chunguang Li and Yufei Zhu and Yaoxing Peng and Lining Sun},
  title = {{Gait parameter fitting and adaptive enhancement based on cerebral blood oxygen information}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2023},
  doi = {10.3389/fnhum.2023.1205858},
}
Figure