A Novel Method to Estimate the Full Knee Joint Kinematics Using Low Cost IMU Sensors for Easy to Implement Low Cost Diagnostics

Mark Versteyhe, Henri De Vroey, Frederik Debrouwere, Hans Hallez, Kurt Claeys

TL;DR

This paper introduces a novel, low-cost method for accurately estimating the full 3D motion of the knee joint (flexion/extension, varus/valgus, and internal/external rotation) using only a few inertial measurement units (IMUs). This approach simplifies diagnostics by offering an easy-to-implement, portable solution that overcomes the cost and complexity limitations of traditional motion capture systems.

What Problem Does This Paper Address?

Assessing knee joint kinematics is crucial for diagnosing injuries, planning surgeries, and guiding rehabilitation. The current gold standard involves expensive and complex optical motion capture systems with reflective markers. While effective, these systems have several drawbacks: they are costly, require specialized laboratory environments, and can struggle to accurately capture certain knee movements like varus/valgus and internal/external rotation due to marker placement variability and camera limitations. This creates a need for more accessible, user-friendly, and cost-effective solutions for clinical diagnostics.

Methods

The core of the proposed method lies in using a Rauch-Tung-Striebel (RTS) filter, a fixed-interval smoother, to estimate the knee joint's full 3D kinematics. Unlike standard Kalman filters that only process data forward in time, the RTS filter performs a backward pass as well, allowing it to leverage future information to refine past estimates and reduce uncertainty.

Key methodological innovations include:

  • Arbitrary Sensor Placement and Quick Calibration: The IMUs can be placed on the upper and lower leg with minimal guidelines, and a short calibration procedure is sufficient, making it highly practical for clinical settings.
  • Simple Dynamic Model: A basic kinematic model of the legs as rigid bodies is employed, incorporating physical constraints between the leg segments and the world reference frame.
  • Combined Sensor Data: The method utilizes both accelerometer and gyroscope data from the IMUs, as magnetic sensors are deemed unreliable due to local field variations.
  • Two-Step Estimation: The process involves first estimating the knee joint's position and orientation relative to the sensor frames, and then using a time-series model with constraints to determine the rigid body motion in a world reference frame.
  • Output Model Constraints: Specific constraints are formulated as output equations for the RTS filter. These include matching sensor acceleration measurements to predicted values, ensuring consistency in joint axis orientation between the upper and lower leg, and limiting drift in the vertical axis rotation.

Datasets

The paper describes two primary evaluation scenarios:

  • Robotic Manipulator (Human Analogue): A robotic arm (ABB IRB 120) was used to simulate controlled knee movements, including large flexion/extension and smaller varus/valgus/internal/external rotations. This served as a "human analogue" to generate ground truth data for validating the IMU-based method.
  • Lunge Movement (Clinical Data): The method was further validated on clinical data from a lunge movement performed by a human subject. This data was captured simultaneously using the proposed IMU system and a traditional optical motion capture system (Optitrack flex 13) for comparison.

Key Insights and Ideas

  • RTS Filter for Enhanced Accuracy: The RTS filter's ability to smooth data in both forward and backward passes significantly reduces uncertainty and improves the accuracy of kinematic estimates, particularly for the smaller varus/valgus and internal/external rotation angles.
  • Robustness to Sensor Placement: The method's independence from precise sensor placement is a major practical advantage, simplifying setup and reducing potential errors introduced by inconsistent mounting.
  • Low-Cost Diagnostic Potential: By leveraging inexpensive IMUs and a streamlined calibration process, the approach democratizes advanced motion analysis, making it accessible for routine clinical diagnostics and rehabilitation monitoring.
  • Addressing Camera System Limitations: The paper highlights that while camera systems excel at estimating flexion/extension, they often struggle with varus/valgus and internal/external rotation due to the geometry of marker placement and the nature of these movements. The IMU method aims to fill this gap.

Key Results

  • Robotic Analogue Validation: The method demonstrated high accuracy when tested on the robotic manipulator, with peak angle errors of approximately 1 degree for roll, pitch, and yaw angles, which is considered more than satisfactory for clinical analysis.
  • Lunge Movement Validation: In the lunge movement test, the IMU-based method accurately identified the main flexion/extension angle, comparable to the camera system. The authors also express confidence in the method's ability to accurately estimate varus/valgus and internal/external rotation, noting that these estimates fall within reasonable motion ranges.

Limitations and Future Work

  • Limitations:
    • The method assumes the patient moves in a relatively straight line; deviations from this can be accommodated but require specific input.
  • Drift of motion around the vertical axis cannot be detected by accelerometer-based sensors alone, although the RTS filter helps mitigate this.
  • The current model is a simple mechanical model; it does not account for dynamic forces or moments.
  • Future Work:
    • Extension to a full-body model for comprehensive biomechanical analysis.
  • Development of a dynamic model to estimate working moments and resulting forces.
  • Further exploration of accommodating more complex movement paths.

Summary updated

Abstract

Traditional motion capture systems are the current standard in the assessment of knee joint kinematics. These systems are, however, very costly, complex to handle, and, in some conditions, fail to estimate the varus/valgus and internal/external rotation accurately due to the camera setup. This paper presents a novel and comprehensive method to infer the full relative motion of the knee joint, including the flexion/extension, varus/valgus, and internal/external rotation, using only low cost inertial measurement units (IMU) connected to the upper and lower leg. Furthermore, sensors can be placed arbitrarily and only require a short calibration, making it an easy-to-use and portable clinical analysis tool. The presented method yields both adequate results and displays the uncertainty band on those results to the user. The proposed method is based on an fixed interval smoother relying on a simple dynamic model of the legs and judicially chosen constraints to estimate the rigid body motion of the leg segments in a world reference frame. In this pilot study, benchmarking of the method on a calibrated robotic manipulator, serving as leg analogue, and comparison with camera-based techniques confirm the method’s accurateness as an easy-to-implement, low-cost clinical tool.
BibTeX
@inproceedings{versteyhe2020novel,
  author = {Mark Versteyhe and Henri De Vroey and Frederik Debrouwere and Hans Hallez and Kurt Claeys},
  title = {{A Novel Method to Estimate the Full Knee Joint Kinematics Using Low Cost IMU Sensors for Easy to Implement Low Cost Diagnostics}},
  booktitle = {Sensors},
  year = {2020},
  url = {https://openreview.net/forum?id=1lItPJeKGV},
}
Figure