We propose a generalized resolution modeling (RM) framework, including extensive task-based optimization,
wherein we continualize the conventionally discrete framework of RM vs. no RM, to include varying degrees of RM.
The proposed framework has the advantage of providing a trade-off between the enhanced contrast recovery by RM and
the reduced inter-voxel correlations in the absence of RM, and to enable improved task performance. The investigated
context was that of oncologic lung FDG PET imaging. Given a realistic blurring kernel of FWHM h (‘true PSF’), we
performed iterative EM including RM using a wide range of ‘modeled PSF’ kernels with varying widths h. In our
simulations, h = 6mm, while h varied from 0 (no RM) to 12mm, thus considering both underestimation and
overestimation of the true PSF. Detection task performance was performed using prewhitened (PWMF) and nonprewhitened
matched filter (NPWMF) observers. It was demonstrated that an underestimated resolution blur (h = 4mm)
enhanced task performance, while slight over-estimation (h = 7mm) also achieved enhanced performance. The latter is
ironically attributed to the presence of ringing artifacts. Nonetheless, in the case of the NPWMF, the increasing intervoxel
correlations with increasing values of h degrade detection task performance, and underestimation of the true PSF
provides the optimal task performance. The proposed framework also achieves significant improvement of
reproducibility, which is critical in quantitative imaging tasks such as treatment response monitoring.
We propose a novel framework of robust kinetic parameter estimation applied to absolute ow quanti cation in dynamic PET imaging. Kinetic parameter estimation is formulated as a nonlinear least squares with spatial constraints problem (NLLS-SC) where the spatial constraints are computed from a physiologically driven clustering of dynamic images, and used to reduce noise contamination. An ideal clustering of dynamic images depends on the underlying physiology of functional regions, and in turn, physiological processes are quanti ed by kinetic parameter estimation. Physiologically driven clustering of dynamic images is performed using a clustering algorithm (e.g. K-means, Spectral Clustering etc) with Kinetic modeling in an iterative handshaking fashion. This gives a map of labels where each functionally homogenous cluster is represented by mean kinetics (cluster centroid). Parametric images are acquired by solving the NLLS-SC problem for each voxel which penalizes spatial variations from its mean kinetics. This substantially reduces noise in the estimation process for each voxel by utilizing kinetic information from physiologically similar voxels (cluster members). Resolution degradation is also substantially minimized as no spatial smoothing between heterogeneous functional regions is performed. The proposed framework is shown to improve the quantitative accuracy of Myocardial Perfusion (MP) PET imaging, and in turn, has the long-term potential to enhance capabilities of MP PET in the detection, staging and management of coronary artery disease.