n.138 – BRAIN 18F-FDG/PET: IMAGE PROCESSING WITH A DEDICATED ALGORITHM (3D-SSP), USING MATRIX WITH DIFFERENT RESOLUTION



Abstract

BACKGROUND-AIM
Alzheimer’s disease is an increasingly popular event in the world’s population. PET imaging with FDG presents characteristic patterns for this disease. 3D-SSP algorithm allows a semiquantitative evaluation automatic and not operator dependent of the PET images for the analysis of patients with Alzheimer’s. GE-CortexID™ software based on this algorithm and is equipped with a database of PET images of normal patients processed with matrix 128×128. The use of matrix to better resolution such as 256×256 matrix are in the normal routine of the current PET/CT tomographs, the purpose of this work was to assess any significant differences in the quantification using in processing matrix in different resolution (128×128 and 256×256).
METHODS
Were studied 37 patients (17 normal and 20 pathological), acquired following the guidelines of AIMN, with tomograph PET/CT: Discovery 690 – GE Healthcare. Were performed image processing simultaneously with 128×128 and 256×256 matrix. Each sequence of images has been reworked with the CortexID™ and the values of Z-score obtained were evaluated in order to calculate the differences between the two matrix used. Subsequently were calculated tolerance values (theory of mean square error).
RESULTS
The differences over the tolerance limits are not homogeneous, but distributed in more areas of the brain, independently that patients are healthy or pathological. In particular the area relative to the caudate nucleus presents many differences and also the Z-score arising from reworks with 256×256 matrix are in 95% of patients, underestimated than the Z-score derived from reworks with 128×128 images. Instead, the values of the area relative to the cerebellum are constantly overestimated in reworks with 256×256 matrix, even if with minor differences.
CONCLUSION
Visual images (axial, sagittal, coronal) processed with 256×256 matrix are qualitatively better. The semiquantitative data obtained instead are discordant respect to reworks with 128×128 matrix and then non-overlapping, even though for a correct analysis of the data it is necessary a higher number of patients.

Scarica PDF