Start Over

Segmentation of gliomas in 18 F-fluorocholine PET/CT. A multiapproach study.

Authors :: García Vicente AM
Pérez-Beteta J
Jiménez Londoño GA
Amo-Salas M
Pena Pardo FJ
Villena Martín M
Borrás Moreno JM
Soriano Castrejón Á
Source :: Revista espanola de medicina nuclear e imagen molecular [Rev Esp Med Nucl Imagen Mol (Engl Ed)] 2019 Nov - Dec; Vol. 38 (6), pp. 362-369. Date of Electronic Publication: 2019 Oct 24.
Publication Year :: 2019
Abstract: Aim: Our aim was two-fold, to study the interobserver agreement in tumour segmentation and to search for a reliable methodology to segment gliomas using <superscript>18</superscript> F-fluorocholine PET/CT.<br />Methods: 25 patients with glioma, from a prospective and non-randomized study (Functional and Metabolic Glioma Analysis), were included.Interobserver variability in tumour segmentation was assessed using fixed thresholds. Different strategies were used to segment the tumours. First, a semi-automatic tumour segmentation was performed, selecting the best SUVmax-% threshold for each lesion. Next we determined a variable SUVmax-% depending on the SUVmax. Finally a segmentation using a fixed SUVmax threshold was performed. To do so, a sampling of 10 regions of interest (ROI of 2.8cm <superscript>2</superscript> ) located in the normal brain was performed. The upper value of the sample mean SUVmax±3 SD was used as cut-off. All procedures were tested and classified as effective or not for tumour segmentation by two observer's consensus.<br />Results: In the pilot segmentation, the mean±SD of SUVmax, SUVmean and optimal SUVmax-% threshold were: 3.64±1.77, 1.32±0.57 and 21.32±8.39, respectively. Optimal SUVmax-% threshold showed a significant association with the SUVmax (Pearson=-0.653, p=.002). However, the linear regression model for the total sample was not good, that supported the division in two homogeneous groups, defining two formulas for predicting the optimal SUVmax-% threshold. As to the third procedure, the obtained value for the mean SUVmax background+3 SD was 0.33. This value allowed segmenting correctly a significant fraction of tumours, although not all.<br />Conclusion: A great interobserver variability in the tumour segmentation was found. None of the methods was able to segment correctly all the gliomas, probably explained by the wide tumour heterogeneity on <superscript>18</superscript> F-fluorocholine PET/CT.<br /> (Copyright © 2019 Sociedad Española de Medicina Nuclear e Imagen Molecular. Publicado por Elsevier España, S.L.U. All rights reserved.)