Optimizing MR-based attenuation correction in hybrid PET/MR using deep learning: validation with a flatbed insert and...

Area of Science:

• Medical Imaging
• Radiology
• Nuclear Medicine

Background:

• Positional mismatches and alignment issues challenge the verification of MR-based attenuation correction (MRAC) in PET/MR imaging.
• Accurate MRAC is crucial for reliable quantitative analysis in hybrid PET/MR scanners.

Purpose of the Study:

• To precisely validate MRAC methods by ensuring accurate MR-CT matching.
• To evaluate the performance of different MRAC techniques, including those using continuous and discrete μ-maps.
• To assess the impact of bone proximity on SUV quantification using MRAC.

Main Methods:

• A validation dataset of 21 patients underwent whole-body [18F]FDG PET/CT followed by [18F]FDG PET/MR, utilizing a flatbed insert for consistent positioning.
• Four MRAC methods were compared against CT-based attenuation correction (CTAC), with synthesized-CTs generated from MR images using a deep learning framework.
• Quantitative analyses included whole-body, ROI, and lesion-level assessments, alongside lesion-distance analysis to evaluate bone proximity effects on SUV.

Main Results:

• The continuous μ-map MRAC method (MRAC-4) showed close alignment with CTAC in joint histogram analysis.
• MRAC-4 effectively minimized bone-induced SUV interference, as indicated by lesion-distance analysis (r=0.01, p=0.8643).
• However, MRAC-4 exhibited greater SUV variability and lower reproducibility in bone-rich regions like the spine and liver compared to discrete μ-map methods.

Conclusions:

• Precise MRAC validation was achieved using a flatbed insert, confirming the potential of MRAC in PET/MR imaging.
• The continuous μ-value MRAC method (MRAC-4) offers superior accuracy and reduces SUV errors related to bone, but requires further refinement for reproducibility in challenging anatomical areas.