A new method to determine probability of malignancy using dynamic breast F-18-FDG PET studies

Objectives To increase specificity for malignancy detection in dynamic breast F-18-FDG PET studies.

Methods Immediately after intravenous administration of 10 mCi F-18-FDG a CT scan on a PET/CT scanner followed by a dynamic breast PET studies were performed. They consisted of 50 frames of 1-minute duration each. Consecutive PET images were nonrigidly registered to the first frame using a finite element method and fiducial skin markers. Nonlinear Time-Activity-Curve (TAC) fitting, based on a realistic two-compartment model, was performed on a voxel-by-voxel basis. Patlak analysis was performed on all fitted TACs. Parametric graphs of correlation between Patlak intercept and Patlak slope for whole breast volume, with every voxel contributing one data point, were created (~500,000 voxels). Voxels contributing to the parametric graph are color-coded to display their physical location (e.g. region of tissue or lesion of origin). In addition, 3D images of spatial distribution of Patlak intercept and slope are obtained registered with the CT image, thus allowing determination of areas with high values of these parameters.

Results After registration and curve-fitting voxels contributing from areas with different metabolic properties occupy well-differentiated and more compact regions with smaller overlap in the obtained parametric graphs. There is a strong correlation between location of area on the parametric graph and probability of malignancy.

Conclusions The proposed parametric graph method might improve specificity of dynamic breast F-18-FDG PET studies for breast cancer detection.