A Novel Marker-Based Registration Method for Simultaneous Preclinical PET/MR with a Non- Stationary PET Detector

Technical Summary: A Novel Marker-Based Registration Method for Simultaneous Preclinical PET/MR with a Non-Stationary PET Detector

Problem Statement
The study addresses the challenge of registering simultaneously acquired Positron Emission Tomography (PET) and Magnetic Resonance (MR) images in a preclinical setting where the PET detector is non-stationary and retrofitted onto an MR scanner. Specifically, the authors utilized a small, non-stationary PET detector (RatCAP) attached to an animal cradle positioning system. This setup presents three primary obstacles to standard registration:

1. Positional Instability: The physical location of the detector relative to the MR magnet isocenter varies with every experimental repositioning of the cradle (with known axial positioning errors), preventing the use of a single, one-time calibration phantom.
2. Spatial Constraints: The inner diameter of the radio-frequency (RF) coil inside the PET detector is too narrow to accommodate dual-modality fiducial markers alongside the imaging subject.
3. Lack of Anatomical Contrast: Relying solely on image intensity-based registration is often insufficient due to the lack of mutual anatomical contrast between PET and MR modalities in many preclinical applications.

Methodology
The authors propose a fully automated, marker-based registration pipeline that decouples the PET image space from the MR image space using a fixed geometric reference within the animal cradle. The method relies on a Zero-Echo-Time (ZTE) MRI sequence capable of imaging the plastic components of the cradle. The workflow involves three specific affine transformations:

1. Universal Calibration ($T_{P \to Cradle}$): A one-time calibration establishes the static relationship between the PET image coordinate system and the cradle coordinate system. This is achieved by simultaneously imaging a dual-modality phantom (stacked syringes filled with $^{18}$F-FDG) and the cradle's nose-cone track (NCT). The NCT was modified with subtractive "nocks" (geometric markers) made of Acrylonitrile Butadiene Styrene (ABS), a material visible in ZTE-MRI. The PET and ZTE images are registered using Advanced Normalization Tools (ANTs) with a Mutual Information metric.
2. Experiment-Specific Calibration ($T_{Lab \to Cradle}$): For each experiment, the location of the cradle within the laboratory space (magnet isocenter) is determined. A short ZTE scan (~2.5 minutes) captures both the subject and the NCT markers. This scan is rigidly registered to a pre-acquired "NCT0" scan (containing only the NCT markers) to determine the cradle's current position relative to the lab space.
3. Sequence-Specific Transformation ($T_{MRI \to Lab}$): This transformation, derived directly from MR image headers, maps the indexed voxels of the specific MRI sequence to the physical laboratory space.

The final registration of PET to MRI is achieved by concatenating these transforms: $x_{MRI} = T_{MRI \to Lab}^{-1} \cdot T_{Lab \to Cradle}^{-1} \cdot T_{P \to Cradle} \cdot x_{PET}$.

Key Contributions

• Novel Hardware Integration: The design and 3D printing of a modified nose-cone track with subtractive geometric markers visible in ZTE-MRI, eliminating the need for space-consuming dual-modality fiducials.
• Automated Workflow: A fully automated registration pipeline that does not require manual intervention, anatomical feature matching, or repeated dual-modality calibration phantoms for every subject.
• Sub-Voxel Precision: The demonstration that accurate registration is possible even when the PET detector moves between experiments, provided the geometric relationship between the detector and the cradle markers remains fixed.

Results
The method was validated through systematic phantom experiments:

• Registration Accuracy: The experiment-specific registration of the cradle markers (NCT) to the laboratory space achieved a spatial error of $\pm 0.03$ mm (x-axis), $\pm 0.06$ mm (y-axis), and $\pm 0.06$ mm (z-axis).
• Coregistration Performance: When applying the full pipeline to PET and MR images of a phantom across nine different cradle positions (including translations and rotations), the resulting coregistration error was below the PET voxel size.
• Quantitative Metrics: The voxel-wise standard deviation across registered PET images, when mapped to a common FLASH MRI space, averaged 2.6% (intensity-normalized, 95th percentile), with a local maximum of 5.4%.
• Efficiency: The required ZTE calibration scan takes approximately 2.5 minutes, making it feasible as a prescan in a standard protocol.

Significance and Claims
The paper claims to present the first demonstration of using image-based registration of ZTE scans to automate the coregistration of simultaneously acquired PET and MR scans in a non-stationary, retrofitted preclinical setup. The authors emphasize that this approach offers a practical, cost-effective solution for PET/MR systems where physical space is limited and detector positioning is not fixed.

The study modestly notes that the validation was performed entirely on phantoms under controlled conditions. The authors acknowledge that future work is required to assess robustness in living animals, where factors such as subject movement and respiration may introduce artifacts affecting the registration steps. The conclusion states that the method is a "practical method" for achieving sub-voxel precision in these specific constrained environments.