Bias atlases for segmentation-based PET attenuation correction using PET-CT and MR
Jinsong Ouyang1, Se Young Chun2, Yoann Petibon3
1Center for Advanced Radiological Sciences, Division of Nuclear Medicine and Molecular Imaging, Massachusetts General Hospital, Boston; Harvard Medical School, Boston.
IEEE Transactions on Nuclear Science
|June 27, 2014
Summary
Tissue segmentation in PET-CT imaging improves attenuation correction accuracy. Four-class segmentation shows varying bias, while three-class segmentation is sufficient for organs like the heart, liver, and kidneys.
Area of Science:
- Medical Imaging
- Radiophysics
- Biomedical Engineering
Background:
- Accurate attenuation correction is crucial for quantitative positron emission tomography (PET) imaging.
- Computed tomography (CT) based attenuation correction can be limited by tissue heterogeneity, particularly in lungs and bones.
- Magnetic resonance (MR) imaging offers potential for improved tissue characterization in hybrid PET-MR systems.
Purpose of the Study:
- To evaluate voxel-wise PET accuracy and precision using tissue segmentation for attenuation correction.
- To assess the impact of different tissue segmentation strategies on PET quantitative accuracy.
- To determine the feasibility of using MR-based fat segmentation for improved attenuation correction.
Main Methods:
- Applied multiple thresholds to CT images of 23 patients for tissue classification.
- Utilized MR fat/in-phase ratio images for fat segmentation in six patients.
- Generated attenuation maps from segmented tissues for PET reconstruction and computed bias images relative to original CT-based reconstruction.
- Created a mean and standard deviation bias atlas from registered bias images.
Main Results:
- Four-class segmentation (air, lungs, fat, other tissues) resulted in root-mean-square error (RMSE) bias of 15.1% (lungs), 4.1% (fat), 6.6% (non-fat soft tissues), and 12.9% (bones).
- Accurate fat identification was achieved using MR fat/in-phase images.
- Three-class segmentation (air, lungs, other tissues) yielded less than 5% standard deviation of bias in the heart, liver, and kidneys.
- Inter- and intra-patient lung density variations contributed equally to the overall standard deviation of bias in the lungs.
Conclusions:
- Four-class tissue segmentation provides improved, though variable, accuracy for PET attenuation correction.
- Three-class segmentation is adequate for achieving low bias variation in key abdominal organs.
- MR-based fat segmentation enhances accuracy, and understanding lung density variations is important for precise PET quantification.
