Abstract
Positron Emission Tomography (PET) traditionally requires subjects to remain stationary, limiting the study of brain activity during natural movement. This paper presents a robotic system designed to facilitate PET imaging of mobile subjects by tracking head motion in real-time. Utilizing a UR5 robot for coarse physical tracking and a high-precision string-encoder system for fine digital motion correction, the system achieves sub-millimeter accuracy (0.51 mm) with a system latency of 80 ms. This hybrid approach allows for high-sensitivity imaging without requiring the subject to bear the weight of the scanner.
Introduction
Standard PET imaging is constrained by the requirement for absolute subject immobility. The weight and bulk of high-sensitivity PET scanners necessitate stationary configurations, which precludes the investigation of neurological processes during locomotion or other natural activities. While wearable "helmet" PET scanners have been developed, they often sacrifice sensitivity to remain lightweight.
To bridge this gap, a robotic motion compensation system is required. Such a system must maintain a heavy imaging ring centered around a moving subject's head. This involves two distinct tasks: coarse motion correction to prevent collisions between the subject and the scanner, and fine motion correction to provide the precise spatial data required for artifact-free image reconstruction.
Methods and System Architecture
The researchers developed a prototype utilizing a dual-robot configuration to simulate and compensate for human movement.
Hardware Configuration
The experimental setup consisted of:
- Subject Simulation: A UR3 robot was programmed to emulate human head trajectories recorded from subjects walking on treadmills.
- Imaging Platform: A UR5 robot carried a mock PET imaging ring.
- Sensing System: A parallel Stewart Platform configuration of six string encoders connected the simulated "head" to the "ring," providing 6-Degree-of-Freedom (DOF) pose data.
- Validation System: To simulate PET data correction, four laser diodes were mounted on the "head" and four cameras on the imaging ring. This allowed the team to mathematically reconstruct the source position based on laser strikes.
Control and Signal Processing
To minimize latency, an FPGA-based interface processed encoder signals at 1 kHz. The UR5 robot was controlled via Cartesian velocity mode. The control algorithm employed a feedforward velocity component combined with a non-linear proportional gain. This non-linear approach allowed the system to ignore minor noise while reacting aggressively to significant movements, maintaining a safe 18 mm clearance between the subject and the scanner.
Experimental Data
The system was tested using a dataset of human head motion trajectories. These trajectories were originally captured at 60 Hz using optical markers during overground and treadmill walking. The data was processed using Savitzky-Golay filters to generate smooth targets for the UR3 robot, ensuring the simulation accurately reflected the velocities and accelerations of a walking human.
Results
The prototype demonstrated the feasibility of robotic motion compensation for PET imaging:
- Latency: Total system latency was measured at 80 ms, falling below the 100 ms threshold identified as necessary for safe operation.
- Accuracy: The string encoder system achieved a static measurement accuracy of 0.5 mm.
- Tracking Performance: During simulated walking, the UR5 maintained the required clearance, with a mean translational tracking error of 5.8 mm in the x-y plane.
- Reconstruction Precision: After applying fine motion correction algorithms to the mock laser-PET data, the residual error was reduced to 0.51 mm. This meets the sub-millimeter precision requirements for high-quality PET image reconstruction.
Discussion and Future Directions
The study highlights the advantages of a hybrid correction strategy. By separating the task into coarse robotic tracking and fine algorithmic correction, the system avoids the need for perfect mechanical rigidity. Furthermore, the use of string encoders proved superior to optical tracking in this high-proximity application due to lower latency and higher robustness.
However, several challenges remain. The UR5 robot used in the prototype cannot support the 15–20 kg weight of a clinical PET ring; future versions will require higher-payload manipulators like the UR20. Additionally, the current reactive controller could be improved with predictive algorithms that leverage the periodic nature of human walking to further reduce effective latency. Finally, operating heavy machinery in close proximity to a human subject requires rigorous safety protocols and further study into human comfort and trust.
Conclusion
This robotic motion compensation method provides a viable path toward high-resolution PET imaging of mobile subjects. By achieving sub-millimeter residual error and low system latency, the system demonstrates that the physical constraints of traditional PET imaging can be overcome through integrated robotic and algorithmic solutions.
Summary updated