A flexible image segmentation prior to parametric estimation

Area of Science:

• Nuclear Medicine
• Medical Imaging
• Computational Biology

Background:

• Positron emission tomography (PET) is crucial for quantitative imaging.
• Accurate parametric imaging requires robust methods for kinetic parameter estimation.
• Existing methods may have limitations in handling complex biological processes.

Purpose of the Study:

• To introduce a flexible method for computing parametric images in PET using spatial and temporal pixel variance.
• To evaluate the accuracy of this new method for brain studies using [(18)F]fluorodeoxyglucose and [(15)O]water.
• To compare kinetic parameter estimation from segmented versus usual images.

Main Methods:

• Developed a method based on spatial and temporal pixel variance for PET parametric imaging.
• Segmented brain images using coefficients of variation and correlation coefficients of neighboring pixels.
• Estimated kinetic parameters using dynamic (DYN) and autoradiographic (ARG) fitting models.

Main Results:

• The proposed method successfully computed parametric images for [(18)F]fluorodeoxyglucose and [(15)O]water brain studies.
• Regional glucose metabolism (rCMRGlc) and blood flow (rCBF) were estimated using both DYN and ARG fitting on segmented and usual images.
• Maximal relative errors were 4% for ARG rCMRGlc, 10% for DYN rCMRGlc, and 17% for DYN rCBF.

Conclusions:

• The pixel variance-based method offers a flexible approach to PET parametric imaging.
• Image segmentation based on pixel variance improves the accuracy of kinetic parameter estimation.
• This method provides reliable quantitative data for brain metabolism and blood flow studies.