MR-Guided Kernel EM Reconstruction for Reduced Dose PET Imaging.

Area of Science:

• Medical Imaging
• Image Reconstruction
• Nuclear Medicine

Background:

• Positron Emission Tomography (PET) image reconstruction is sensitive to Poisson noise.
• Reduced acquisition times or lower injected doses exacerbate noise issues in PET data.
• Kernel Expectation Maximization (KEM) is a novel technique for noise reduction in PET images.

Purpose of the Study:

• To evaluate the effectiveness of the kernel method with MR-derived kernels for noise reduction in PET imaging.
• To assess the potential for dose reduction in PET scans using the kernel method.
• To compare the performance of KEM against Maximum Likelihood Expectation Maximization (MLEM) for low-count PET datasets.

Main Methods:

• Utilized MR-derived kernels within the kernel expectation maximization (KEM) algorithm.
• Reconstructed PET data using both KEM and MLEM algorithms.
• Compared reconstruction quality and quantitative metrics for varying data counts, simulating reduced injected doses.
• Evaluated whole-brain voxel-level error and regional metrics (coefficient of variation, mean values).

Main Results:

• KEM demonstrated superior performance over MLEM for low-count PET datasets, yielding visibly clearer reconstructions.
• KEM reconstruction using 10% of the data showed comparable voxel-level error to MLEM using 100% of the data (simulated).
• At reduced dose levels, KEM achieved a lower coefficient of variation and more accurate mean values compared to MLEM.
• Potential over-smoothing of PET-specific features was observed as a trade-off with the kernel method.

Conclusions:

• The kernel method, particularly with MR-derived kernels, offers significant noise reduction and enables substantial dose reduction in PET imaging.
• KEM provides improved image quality and quantitative accuracy for low-count PET data compared to MLEM.
• Further clinical validation is necessary to determine optimal dose reduction levels while preserving diagnostic utility.