One registration multi-atlas-based pseudo-CT generation for attenuation correction in PET/MRI

Area of Science:

• Medical Imaging
• Radiology
• Image Processing

Background:

• Atlas-based segmentation is crucial for whole-body bone identification in MRI.
• Accurate attenuation maps are essential for PET/MRI attenuation correction.
• Existing methods involve computationally intensive registration processes.

Purpose of the Study:

• To develop and evaluate a novel one-registration multi-atlas (ORMA) approach for pseudo-CT generation.
• To optimize parameters for atlas-based whole-body bone segmentation from MRI.
• To improve the efficiency of generating attenuation maps for PET/MRI.

Main Methods:

• Implemented a one-registration strategy where only one atlas is registered online.
• Utilized pre-computed transformations for remaining atlases to align with the target image.
• Compared ORMA with direct registration methods (VWW and arithmetic averaging) using PET/MRI data.

Main Results:

• Voxel-wise weighting (VWW) atlas fusion achieved a Dice similarity of 0.82 ± 0.04 for bone extraction.
• ORMA-VWW yielded a Dice similarity of 0.76 ± 0.05, showing a modest accuracy decrease.
• Quantitative PET analysis demonstrated good correlation with CT-based attenuation correction, with improved SUV bias.

Conclusions:

• The ORMA approach effectively generates pseudo-CT attenuation maps with reduced computational time.
• A reasonable compromise between computing time and quantitative accuracy was achieved.
• The method shows potential for efficient PET attenuation correction in clinical settings.