Abstract
43
Objectives Dynamic F-18 FET PET is increasingly used in the diagnosis of brain tumors. Taking into account pixel-wise information of the dynamic data, we applied a two-tissue compartment model. A new algorithm of hierarchical pharmacokinetic modeling (HPM) was developed in order to improve the quality of parametric maps, as fitting of nonlinear models suffers from local minima.
Methods Our algorithm of HPM aims to improve the parametric image estimation by refining the initial values and fitting boundaries hierarchically to reduce the local minima of nonlinear fitting. Then pixel-wise pharmacokinetic modeling was performed with refined initial values and fitting boundaries. The algorithm was tested on computational simulations. Root mean square error (RMSE) was estimated and compared with the results of direct pixel-wise modeling (PMOD). The algorithm was further tested on 8 patient datasets (4 with low grade glioma (LGG) and 4 with high grade glioma (HGG)) obtained from 40 min. dynamic F-18 FET PET/MRs comparing visual and quantitative features of the calculated FET parameters (K1-k4, VB) in the anatomically defined (3D T2 FLAIR datasets) tumors.
Results Compared to the results of direct pixel-wise modeling, our algorithm reduced the RMSE (50% to 92% reduction) of the estimated parametric images. For patient data, HGG patients had higher k3 and k4 values in tumors compared to LGG patients. Strong contrast and increased k4 has been observed at the location of HGG tumor compared to normal tissue as well as to the tumor location of patients with LGG tumor. Rather low contrast between HGG and LGG tumor and normal tissue has been observed in k3 images.
Conclusions We show that HPM based on dynamic F-18 FET PET/MR is clinically feasible and has the potential to assist the grading of glioma. In order to validate our preliminary parametric findings a histopathological validation study based on F18-FET PET/MR image-guided stereotactic biopsies in glioma patients is currently underway.