Time-resolved PET imaging using a joint dynamic reconstruction and motion estimation framework (DREME-PET)

Area of Science:

• Medical Imaging
• Radiotherapy
• Nuclear Medicine

Background:

• Positron emission tomography (PET) is crucial for image-guided radiotherapy, requiring accurate tumor localization and accounting for respiratory motion.
• Time-resolved PET imaging is desirable for capturing dynamic anatomical variations but faces challenges due to low photon counts and ill-posed reconstruction problems.

Purpose of the Study:

• To introduce DREME-PET, a dynamic reconstruction and motion estimation framework for time-resolved PET imaging and respiratory motion tracking.
• To enable accurate volumetric PET imaging directly from low-count PET list-mode data.

Main Methods:

• DREME-PET reconstructs time-resolved PET images and generates a patient-specific motion model from conventional PET scans.
• It utilizes spatial implicit neural representation for reference PET images and a low-rank motion model with B-spline interpolants and a CNN encoder for motion estimation.
• The framework was trained with data augmentation to generalize to varying motion amplitudes and count statistics.

Main Results:

• DREME-PET achieved accurate time-resolved PET reconstruction and motion modeling, enabling real-time predictions at 28 ms/frame.
• Simulation studies showed a 14.1% image contrast error and 1.7 mm tracking error.
• Phantom studies demonstrated a 7.4% image contrast error and 0.5 mm tracking error, while patient studies showed >0.91 correlation and 3.6% image contrast error.

Conclusions:

• DREME-PET enables accurate time-resolved PET reconstruction and motion modeling from pre-treatment scans.
• The framework facilitates real-time PET imaging and motion tracking using minimal data, enhancing PET-guided adaptive radiotherapy.