In quantitative positron emission tomography (PET) brain studies, the temporal dynamics of the radiopharmaceutical are usually analyzed separately for different brain structures. In a clinical environment, the delineation of brain structures is still often performed manually by human experts. In this study, we concentrate on automatic segmentation of the striatal brain structures (caudate, posterior and anterior putamen and ventral striatum) from the binding potential (BPND) images derived based on [11C]-raclopride studies. Previously, a method for the automatic segmentation of the striatal structures was proposed for the ECAT high resolution research tomograph (HRRT, CTI PET Systems, Knoxville, TN, USA) BPND images. The method is based on clustering the affinity matrix (containing the features as intensity values, spatial connectivity and distance) of the striatum which is extracted by using a deformable surface model. In clustering, the method uses weighted kernel k-means algorithm. In this study, we evaluate the segmentation method with a test-retest dataset. We studied the segmentation differences between the analytical (3D-RP) and iterative (3D-OPOSEM) reconstructions of the ECAT HRRT data. In addition to visual comparisons, for different reconstruction methods, normalized absolute differences (NAD) between the segmented regions of test-retest BPND images were calculated. We observed that NAD values were within acceptable limits in most cases. However, the ventral striatum segmentation failed to some extent. Furthermore, it is obvious that robustness of this kind of brain structure extraction methods should be tested with various reconstruction methods.
[1]
Ulla Ruotsalainen,et al.
Segmentation of Striatal Brain Structures from High Resolution PET Images
,
2009,
Int. J. Biomed. Imaging.
[2]
A. Lammertsma,et al.
Simplified Reference Tissue Model for PET Receptor Studies
,
1996,
NeuroImage.
[3]
U. Ruotsalainen,et al.
Automatic Extraction of Caudate and Putamen in [$^11$C] Raclopride PET Using Deformable Surface Models and Normalized Cuts
,
2006,
IEEE Transactions on Nuclear Science.
[4]
Inderjit S. Dhillon,et al.
Weighted Graph Cuts without Eigenvectors A Multilevel Approach
,
2007,
IEEE Transactions on Pattern Analysis and Machine Intelligence.
[5]
Uygar Tuna,et al.
Interpolation for the gap-filling of the HRRT PET sinograms by using the slices in the direction of the radial samples
,
2009,
2009 IEEE Nuclear Science Symposium Conference Record (NSS/MIC).
[6]
Uygar Tuna,et al.
Gap-Filling for the High-Resolution PET Sinograms With a Dedicated DCT-Domain Filter
,
2010,
IEEE Transactions on Medical Imaging.