Data-Driven Motion Detection and Event-by-Event Correction for Brain PET: Comparison with Vicra.

Area of Science:

• Neuroimaging
• Medical Physics
• Radiochemistry

Background:

• Head motion in PET studies degrades image quality and leads to inaccurate kinetic modeling.
• Current methods like frame-based image registration (FIR) and hardware-based motion tracking (HMT) have limitations in clinical applicability and correction scope.

Purpose of the Study:

• To develop and evaluate a novel data-driven motion correction framework using PET raw data to overcome limitations of existing methods.
• To assess the performance of the centroid of distribution (COD) algorithm against FIR and HMT for head motion correction in brain PET.

Main Methods:

• A data-driven algorithm, centroid of distribution (COD), was developed to detect head motion by analyzing event data in 1-second intervals.
• Frames identified with motion were reconstructed without attenuation correction and rigidly registered to a reference frame.
• The COD framework was applied to 23 human dynamic PET datasets (18F-FDG and 11C-UCB-J) and compared with FIR and Vicra HMT.

Main Results:

• The COD method demonstrated superior accuracy, with SUV differences of 1.0% ± 3.2% for 18F-FDG and 3.7% ± 5.4% for 11C-UCB-J compared to HMT.
• For dynamic studies, COD yielded kinetic parameter differences of 3.6% ± 10.9% for Ki (18F-FDG) and 3.7% ± 5.2% for VT (11C-UCB-J) versus HMT.
• No motion correction (NMC) and FIR showed significantly larger errors, with NMC yielding -15.7% ± 12.2% and FIR -4.7% ± 6.9% for SUV differences.

Conclusions:

• The proposed COD-based data-driven motion correction method effectively addresses limitations of FIR and HMT.
• COD achieves comparable or superior performance to gold-standard HMT, offering a promising solution for accurate brain PET imaging.
• This method enhances the reliability of tracer kinetic modeling in clinical PET studies.