Assessment of brain surface extraction from PET images using Monte Carlo simulations

In this paper, we evaluate quantitatively the performance of the fully automatic deformable model with dual surface minimization (DM-DSM) method for brain surface extraction from positron emission tomography (PET) images with Monte Carlo simulated data. In addition, we cross validate the DM-DSM method with a method based on MRI-PET registration for PET brain delineation. For this, the automated image registration (AIR) algorithm is combined with the anatomical brain surface extractor (BSE) algorithm. Two radiopharmaceuticals were considered: C-11-Raclopride and F-18-FDG. The success of the two methods was quantified by measuring the similarity between the extracted and the true brain volume. Also local differences between the extracted and the true brain surfaces were measured. With FDG, the DM-DSM method yielded brain surfaces of high accuracy and they were more accurate than with the image registration based method. With Raclopride, the accuracy of the DM-DSM method was slightly lower than with FDG and, on the average, similar to the accuracy of the image registration based method. However with Raclopride, maximal local differences between true and extracted surfaces were found to be greater with DM-DSM. In addition, preliminary experiments with images containing simulated pathology were done and the performance of DM-DSM was excellent in these experiments. To summarize, we found that the DM-DSM method can reliably extract brain surfaces of high accuracy from PET images.