MCP-Net: Inter-frame Motion Correction with Patlak Regularization for Whole-body Dynamic PET
Xueqi Guo1, Bo Zhou1, Xiongchao Chen1
1Yale University, New Haven, CT 06511, USA.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|March 11, 2024
Summary
MCP-Net corrects inter-frame patient motion in dynamic PET imaging by optimizing tracer kinetics. This deep learning framework improves spatial alignment and quantitative accuracy of parametric images.
Area of Science:
- Medical Imaging
- Nuclear Medicine
- Artificial Intelligence
Background:
- Inter-frame patient motion in dynamic PET causes spatial misalignment, degrading parametric image quality.
- Current deep learning methods often overlook tracer kinetics in motion correction.
Purpose of the Study:
- To introduce MCP-Net, a novel framework for inter-frame motion correction in dynamic PET.
- To improve parametric imaging by directly optimizing Patlak fitting error.
Main Methods:
- MCP-Net integrates motion estimation (3D U-Net with ConvLSTM), image warping, and analytical Patlak fitting.
- A Patlak loss term, including mean squared percentage fitting error, is incorporated.
- Parametric images are generated using standard Patlak analysis post-correction.
Main Results:
- MCP-Net effectively corrects residual spatial mismatch in dynamic PET frames.
- Improved spatial alignment of Patlak Ki/Ve images was observed.
- Normalized fitting error was significantly reduced compared to benchmarks.
Conclusions:
- MCP-Net enhances quantitative accuracy in dynamic PET by utilizing tracer dynamics.
- The framework shows potential for improving overall PET image analysis and diagnostic capabilities.
