Gait Event Detection and Travel Distance Using Waist-Worn Accelerometers across a Range of Speeds: Automated Approach

Albara Ah Ramli, Xin Liu, Kelly Berndt, Chen-Nee Chuah, Erica Goude, Lynea B. Kaethler, Amanda Lopez, Alina Nicorici, Corey Owens, David Rodriguez, Jane Wang, Daniel Aranki, Craig M. McDonald, Erik K. Henricson

TL;DR

This paper introduces a novel, automated approach using a single waist-worn accelerometer to accurately detect gait events and estimate travel distance across various speeds. This method is particularly promising for monitoring individuals with Duchenne Muscular Dystrophy (DMD) who have atypical gaits, offering a more accessible and less intrusive alternative to traditional clinical assessments.

What Problem Does This Paper Address?

Accurately measuring temporospatial gait features (like step count, length, and speed) is crucial for assessing mobility, especially in populations with conditions like Duchenne Muscular Dystrophy (DMD). Individuals with DMD experience progressive loss of ambulatory function, leading to altered gait patterns that make standard measurement techniques challenging. Traditional methods often rely on laboratory settings with specialized equipment (like force plates) or require direct clinical observation, which are impractical for community-based, long-term monitoring. Existing wearable solutions, like smartphone pedometers, often struggle with accuracy, particularly for individuals with atypical gaits or across a wide range of walking and running speeds. This paper aims to bridge this gap by developing an automated, unsupervised system that can reliably capture these gait metrics using a readily available sensor.

Methods

The core of this work is a machine learning (ML)-based signal processing method that combines clinical observation with regression techniques. The approach involves:

  • Calibration: A crucial step involves calibrating the system to each individual's unique gait characteristics. This is achieved by having participants perform standardized walking/running tests at five different speeds (SC-L1 to SC-L5).
  • Step Detection: Raw accelerometer data from the anteroposterior axis (identified as most informative) is filtered to smooth out noise. Peaks in the filtered signal are identified as indicative of individual steps.
  • Step Length Estimation: A personalized, non-linear regression model is built for each participant. This model maps the peak acceleration values observed during calibration to corresponding step lengths. This individualized approach accounts for variations in gait patterns between individuals.
  • Distance Estimation: Once the step lengths are estimated for each detected step, the total distance traveled is calculated by summing these individual step lengths.
  • Gait Event Detection (Initial Contact - IC): The system identifies the midpoint between two peaks in the filtered signal to approximate toe-off events. Peaks in the original acceleration signal within each step duration are then used to pinpoint the maximum peak value, corresponding to the initial contact (IC) event.

This method aims to be "unsupervised" in its application, requiring only an initial calibration phase.

Datasets

The study utilized data collected from 30 participants: 15 children with Duchenne Muscular Dystrophy (DMD) and 15 typically developing (TD) peers. These participants underwent supervised testing across various gait activities:

  • Speed Calibration Tests (SC-L1 to SC-L5): Five distinct speeds used for individual calibration.
  • 6-Minute Walk Test (6 MWT): A standard test of endurance.
  • 100-meter Fast Walk/Jog/Run (100 MRW): A test of faster locomotion.
  • Free Walk (FW): Unstructured walking within a defined area.

Ground-truth data was collected through video observation, serving as the benchmark for evaluating the system's accuracy. The paper also compares its performance against the built-in pedometer function of an iPhone.

Key Insights and Ideas

  • Individualized Calibration is Key: The most significant insight is the necessity of individual calibration. Gait patterns are highly personal, and a one-size-fits-all approach fails, especially for atypical gaits. The regression model trained on individual speed data proved highly effective.
  • Waist-Worn Accelerometer Sufficiency: The study demonstrates that a single accelerometer placed at the waist, near the body's center of mass, is sufficient for accurate gait analysis across a range of speeds, eliminating the need for more complex sensor setups or ground reaction force measurements.
  • DMD Gait Patterns are Differentiable: The gait pattern representation (Figure 5) visually highlights the distinct differences in acceleration patterns between TD individuals and those with DMD, underscoring the need for specialized analysis methods.
  • Beyond Pedometer Accuracy: The results strongly suggest that this ML-based approach significantly outperforms standard pedometer functions, particularly for individuals with gait impairments, opening doors for more reliable remote monitoring.

Key Results

The proposed Walk4Me system demonstrated high accuracy in estimating gait parameters:

  • Step Counts: Mean percentage error of 1.49% (7.04% SD) for combined 6 MWT, 100 MRW, and FW tasks.
  • Distance Traveled: Mean percentage error of 1.18% (9.91% SD) for combined 6 MWT, 100 MRW, and FW tasks.
  • Step Length: Mean percentage error of 0.37% (7.52% SD) for combined 6 MWT, 100 MRW, and FW tasks.
  • Correlation: Strong correlations (Pearson's r = -0.9929 to 0.9986, p < 0.0001) were observed between predicted and ground-truth values for step counts, distance, and step length.
  • Comparison to iPhone Pedometer: The Walk4Me system showed significantly lower error rates across all metrics compared to the iPhone's built-in pedometer, which exhibited average errors of 48.46% for step counts, 42.23% for distance, and 46.40% for step length.

Limitations and Future Work

  • Calibration Time: While the system is automated for data collection, it requires an initial calibration phase involving five different speed tests. Future work could explore reducing this calibration burden.
  • Generalizability: The study focused on children with DMD and TD peers. Further validation is needed for other populations with different gait disturbances or age groups.
  • Unstructured Environments: While the "Free Walk" (FW) condition aimed to simulate less structured environments, extensive testing in highly variable, real-world settings (e.g., uneven terrain, crowded spaces) would be beneficial.
  • Sensor Placement: The study used a waist-worn accelerometer. Investigating the impact of slight variations in sensor placement or the use of other body locations could be explored.

The authors suggest that this method can be extended to community settings, enabling more frequent and convenient gait data collection, which is vital for monitoring disease progression and treatment efficacy.

Summary updated

BibTeX
@inproceedings{ramli2024gait,
  author = {Albara Ah Ramli and Xin Liu and Kelly Berndt and Chen-Nee Chuah and Erica Goude and Lynea B. Kaethler and Amanda Lopez and Alina Nicorici and Corey Owens and David Rodriguez and Jane Wang and Daniel Aranki and Craig M. McDonald and Erik K. Henricson},
  title = {{Gait Event Detection and Travel Distance Using Waist-Worn Accelerometers across a Range of Speeds: Automated Approach}},
  booktitle = {Sensors},
  year = {2024},
  url = {https://openreview.net/forum?id=9HYXk5CUBS},
}
Figure