Evaluation of a Motion Measurement System for PET Imaging Studies

Junxiang Wang, Ti Wu, Iulian Iordachita, Peter Kazanzides

Abstract

Positron Emission Tomography (PET) imaging of the brain traditionally requires subjects to remain stationary, limiting the study of neurological processes during natural movement. To enable imaging of walking subjects, a robotic system must suspend a heavy PET detector ring (15–20 kg) around the head. This paper evaluates a 6-degree-of-freedom (6-DOF) mechanical tracking system using string encoders in a Stewart platform configuration. The system provides high-frequency (1 kHz) feedback with sub-millimeter accuracy, achieving translation RMS errors below 0.3 mm. These results demonstrate the feasibility of using mechanical tracking to drive low-latency robotic control loops for mobile PET applications.

Introduction

Positron Emission Tomography (PET) is a critical modality for imaging deep brain structures and neurochemical activity. However, the requirement for strict head immobilization prevents the study of brain function during dynamic activities such as gait and balance. While portable PET "helmets" have been proposed, high-sensitivity detectors are too heavy (15–20 kg) for a human to wear safely.

A potential solution involves a robotic arm that supports the weight of the PET ring, moving it in synchrony with the subject's head. This requires a tracking system that is not only accurate but also provides high-frequency updates to maintain the safety and stability of the robot. While optical tracking systems offer high precision, they often suffer from latency (30–60 Hz) and line-of-sight occlusions. This study evaluates a mechanical alternative based on string encoders designed to overcome these limitations.

Methods

System Architecture

The proposed tracking system utilizes a Stewart platform configuration, a parallel robotic structure known for its high stiffness and precision. Six string encoders are mounted between the base (the PET imaging ring) and a lightweight safety helmet worn by the subject. As the subject moves, the strings extend or retract, and the encoders measure these changes in length.

Kinematics and Calibration

The system's pose is determined through forward kinematics. While inverse kinematics (calculating string lengths from a known pose) is straightforward, forward kinematics (calculating pose from string lengths) requires an iterative numerical approach. The researchers implemented an algorithm that typically converges within 3–4 iterations, making it suitable for real-time applications.

During initial testing, a systematic error was identified: the strings did not pass through the geometric center of their guide holes, instead resting against the edges. To compensate for this physical reality, the team identified a 3 mm "string length offset" through empirical testing, which was then integrated into the kinematic model to improve accuracy.

Experimental Validation

To validate the system, the helmet was mounted to a UR3 industrial robot. The UR3 served as the ground-truth reference, with its accuracy verified by a physical dial indicator. The researchers collected 66 data points across all six degrees of freedom (X, Y, Z, Roll, Pitch, and Yaw) within a range of ±10 mm and ±10 degrees to compare the string encoder measurements against the robot's commanded positions.

Results

The mechanical tracking system demonstrated high precision across all dimensions, meeting the 0.5 mm accuracy threshold required for fine motion correction in PET imaging.

Translation Accuracy

The Root-Mean-Square (RMS) errors for translations were:

  • X-axis: 0.223 mm
  • Y-axis: 0.257 mm
  • Z-axis: 0.193 mm

Rotation Accuracy

Rotation errors were slightly higher but remained within acceptable limits for the intended application:

  • Roll: 0.241°
  • Pitch: 0.529°
  • Yaw: 0.377°

The higher error margins in Pitch and Yaw were attributed to potential inaccuracies in the initial setting of the robot's Tool Center Point (TCP) rather than inherent flaws in the encoder system.

Discussion

Frequency and Latency

The most significant advantage of the string encoder system is its 1 kHz sampling frequency. In robotic control, low latency is vital for safety when a heavy payload is moving in close proximity to a human. The high-speed feedback provided by this mechanical system allows for a much tighter control loop than standard optical systems.

Practical Implementation

The discovery of the "hole center" deviation highlights the importance of reconciling CAD models with physical friction and geometry. By applying a mathematical offset rather than redesigning hardware, the researchers demonstrated a pragmatic approach to system calibration. However, the authors note that for a clinical deployment, a redundant sensor fusion approach—combining mechanical, optical, and inertial sensors—would be ideal to ensure safety and robust performance.

Conclusion and Future Work

This evaluation confirms that a Stewart platform-based string encoder system can provide the sub-millimeter accuracy and high-frequency feedback necessary for mobile PET imaging. Future research will focus on testing the system with complex, multi-axial movements derived from real human gait data. Additionally, the team aims to perform a full kinematic calibration to further reduce error and integrate the tracker with a high-payload robot capable of supporting a full-scale PET imaging ring.

Summary updated

BibTeX
@inproceedings{wang2022evaluation,
  author = {Junxiang Wang and Ti Wu and Iulian Iordachita and Peter Kazanzides},
  title = {{Evaluation of a Motion Measurement System for PET Imaging Studies}},
  booktitle = {ISMR},
  year = {2022},
  doi = {10.1109/ismr48347.2022.9807554},
}
Figure