Magnetic resonance-based attenuation correction for PET/MR hybrid imaging using continuous valued attenuation maps.

Area of Science:

• Medical Imaging Physics
• Radiological Sciences
• Biomedical Engineering

Background:

• Accurate attenuation correction is crucial for quantitative Positron Emission Tomography (PET) imaging.
• Generating attenuation maps (μ-maps) from Magnetic Resonance (MR) data in hybrid PET/MR systems is challenging due to signal limitations in cortical bone and air.
• Standard MR sequences struggle to differentiate bone and air, hindering precise attenuation correction.

Purpose of the Study:

• To develop and assess a continuous-valued, CT-like attenuation map (μ-map) derived from dedicated MR sequences.
• To compare the PET quantification accuracy of the MR-derived μ-map against CT-based μ-maps and segmentation-based methods for head imaging.

Main Methods:

• Acquisition of 3D Dixon and ultrashort echo time MR sequences on a Siemens 3-T Biograph mMR PET/MR system.
• Utilized an epsilon-insensitive support vector regression (ε-SVR) technique for pseudo-CT training on patient data.
• Compared four μ-maps: scaled CT (μ-map CT), segmented CT without bone (μ-map no bone), segmented CT with bone (μ-map bone), and MR-predicted (μ-map MR).

Main Results:

• The proposed MR-derived μ-map achieved a mean absolute error of 2.40% for the whole brain and 2.16% near cortical bone.
• Segmentation-based methods showed higher errors: 10.15% (no bone) and 3.96% (with bone) for the whole brain.
• The developed model demonstrated potential for predicting pseudo-CTs in other anatomical regions using only MR information.

Conclusions:

• Continuous-valued attenuation maps can be generated from MR sequences, significantly improving PET quantification accuracy for head studies.
• The MR-derived μ-map approach, with a <2-minute acquisition, shows promise for clinical PET/MR applications.
• Further validation is needed for attenuation correction in body regions beyond the head.