Introduction
Weight-bearing cone-beam computed tomography (WBCT) has emerged as a critical diagnostic tool for orthopedic conditions, particularly osteoarthritis. Unlike traditional supine CT, WBCT allows clinicians to observe the knee joint under the physiological stress of a patient's own body weight. However, the requirement for the patient to remain perfectly still while standing is a significant hurdle. Involuntary "body sway" during the scan leads to motion-corrupted projections, resulting in streaking artifacts, blurring, and double contours that can render the 3D reconstruction clinically unusable.
Existing motion compensation strategies often rely on 2D-3D registration or the attachment of metallic fiducial markers to the patient's skin. While effective, these methods are either computationally intensive or physically cumbersome for the clinician and patient. This paper explores a more practical alternative: using a small, lightweight Inertial Measurement Unit (IMU)—similar to the sensors found in modern smartphones—to track and correct for patient movement in real-time.
Methodology
The researchers conducted a comprehensive simulation study to evaluate the efficacy of IMU-based motion compensation. The workflow integrated real-world human kinematics with high-fidelity numerical phantoms.
Data Acquisition and Modeling
To ensure realistic motion profiles, the study utilized a high-precision Vicon optical tracking system to record the movements of seven healthy subjects performing squats at 30° and 60° flexion angles. These recorded trajectories were then used to animate a biomechanical model in OpenSim and a numerical knee phantom (XCAT). This setup allowed for the generation of synthetic X-ray projections corrupted by authentic human motion.
IMU Simulation and Processing
A virtual IMU was simulated on the shank (lower leg) of the model to generate acceleration and angular velocity signals. The proposed compensation pipeline involved several critical stages:
- Gravity Removal: Distinguishing the constant pull of gravity from the dynamic acceleration of the leg.
- Integration: Converting the raw sensor data into 3D positional and orientational changes over time.
- Coordinate Transformation: Mapping the sensor's local coordinate system into the global coordinate system of the CT scanner.
- Motion-Compensated Reconstruction: Adjusting the CT projection matrices during the reconstruction process to account for the measured movement, effectively "undoing" the motion blur.
Results
The IMU-based method was benchmarked against uncorrected scans and the current "gold standard" marker-based tracking. The results were evaluated using the Structural Similarity Index (SSIM) and Root Mean Squared Error (RMSE).
- Quantitative Performance: The proposed method improved SSIM by 13–21% compared to motion-corrupted reconstructions, achieving an average value of 0.99. Furthermore, the RMSE was reduced by 68–70%. Notably, the IMU-based approach outperformed the marker-based method in certain scenarios, which saw a 57% improvement in RMSE.
- Qualitative Assessment: Visual analysis of the reconstructed volumes showed a dramatic reduction in artifacts. The "double edges" and blurring characteristic of standing CT scans were largely eliminated, producing images comparable in quality to stationary scans.
Discussion and Key Insights
A primary insight of this research is the effectiveness of treating the lower leg as a rigid body. Although the knee joint moves non-rigidly (with the thigh and shank moving independently), a single IMU placed on the shank provided sufficient data to significantly enhance image quality.
Compared to metallic markers, IMUs offer several clinical advantages. They are non-invasive, do not interfere with the X-ray path (avoiding the "streak" artifacts markers can cause), and are significantly easier to deploy in a fast-paced clinical environment. The study demonstrates that IMUs can achieve performance parity with marker-based systems while offering a more streamlined workflow.
Limitations and Future Work
As a simulation-based study, several factors remain to be addressed for clinical translation:
- Sensor Noise: Real-world IMUs are subject to noise and "drift" (accumulated error), which were not fully modeled in this initial study.
- Initial State Estimation: The study assumed the sensor's initial position and velocity were known. Future iterations will need to estimate these parameters, potentially by tracking the sensor in the initial X-ray projections.
- Multi-Sensor Arrays: To account for the non-rigid motion of the thigh, the authors suggest that a dual-sensor setup (one on the shank and one on the thigh) could further refine the compensation.
Conclusion
This research demonstrates that IMU-based motion compensation is a viable and high-performing solution for weight-bearing CT of the knee. By providing a practical alternative to cumbersome marker-based systems, this technology has the potential to make high-quality, weight-bearing diagnostic imaging more accessible and reliable in orthopedic practice.
Summary updated