Feasibility of a brainstem and diencephalon atlas in FDOPA PET / MRI imaging

Objectives: FDOPA cerebral PET is becoming increasingly important in the diagnosis of neurodegenerative diseases. Only striatal uptake is analyzed, the brainstem and the diencephalon being ignored although FDOPA is significantly metabolized by many of their nuclei. Their analysis is very difficult due to the poor visualization on CT, preventing the identification of its structures in PET / CT. In addition, the position of the stem with respect to the cortex varies between patients. This prevents accurate registration of MRI or PET images on cerebral atlases. The use of a hybrid PET / MRI camera overcomes this problem, allowing the registration of images of both modalities into a predefined model. In this context, our objective was to constitute an atlas of FDOPA uptake in the brainstem and diencephalon from images acquired in PET / MRI: the registration of the images on a reference model was carried out in MRI and then applied to PET images, the validation of the results taking place mainly in PET.

Methods: We retrospectively selected 35 patients who underwent FDOPA PET / MRI: 15 considered as positive for dopaminergic denervation (clinically suspected parkinsonian patients with visually decreased putaminal activity), 20 considered as normal (the control group). After reconstruction of the tomographic sections, the T1 images of each patient were registered on the T1 images of a reference model where several anatomical regions had been previously identified. The registration was carried out in a progressive way, in non-deformable and then deformable mode, after creation of a mask including the brainstem, the diencephalon and the striatum. The registration parameters were then applied to the FDOPA images acquired 90 min after injection. The choice of the optimal method of registration was based on the analysis of the dispersion of the activity in the summed image of the control group. A score was created, multiplying each SUV of each voxel of a predefined area by the spatially corresponding SUV in the model. Eight areas of the brainstem were used to create the final score. We finally investigated the correlation between the scores of 70 deformable registrations and the ability of specific brain regions to discriminate, using those registrations, the parkinsonian from the normal patients.

Results: The whole processing takes about 4 minutes to be complete. Visually the images of the brainstem, diencephalon and striatum are correctly replaced after registration in all patients, without deformation. No manual correction was needed. A significant inverse correlation (Pearson’s coefficient; p-values less than 0,0001) was found between the 70 scores and the corresponding p-values of a Student t-test comparing the control and patient groups in 5 non-striatal regions: the epiphysis, the hypothalamus, the pulvinar, the raphe nuclei and the substantia nigra. Furthermore, using the registration method with the highest score, two of those regions (the substantia nigra and the hypothalamus) exceeded the level of signification. Those statistical characters were also observed in well validated regions in Parkinson’s disease, the caudate and the putamen.

Conclusions: Using the hybrid PET / MRI technique, it is possible to reliably achieve an automatic analysis of FDOPA uptake in brainstem and diencephalon nuclei. The method is independent of the injected tracer, which broadens its scope.