Towards optimal model-based partial volume effect correction in oncological PET imaging

Quantification of tumor uptake in nuclear medicine requires the assessment of the tumor size. We propose a method based on simple ellipsoidal models which only require the reconstructed image data and scanner physical parameters. The method uses simple image and system models but is theoretically not sensitive to the reconstructed image sampling, provided that the local resolution parameters of the scanner are known. Interpolation is thus avoided and the degradation model is computed analytically. The proposed method uses the property that the blurring effect of the scanner spreads the activity around the lesion location and in the neighbor-voxels-even for sub-voxel sized lesions-and attempts to use this limited information to accurately restore lesion parameters. We propose solutions to avoid bias and reduce computation time. The uptake of simple simulated tumor objects could be restored and the influence of lesion's shape and neighborhood was investigated on real patient data for which convergence of the algorithm was demonstrated. This approach is a contribution towards in-depth investigation of reconstruction/post-reconstruction-based models for tumor activity restoration based on more complex image and system models.