Abstract
2053
Objectives Surgical resection of cerebral glioma requires highly reproducible size measurement based on preoperative radiological imaging. 11C methionine (MET) PET describes glioma expansion more precisely than MRI, but available automatic methods fail to contour the tumor due to heterogeneity of MET uptake in background brain tissues. The purpose of this study was to investigate reliability of adaptive region-growing (AR) method, a semi-automatic contouring algorism, in comparison with conventional visual thresholding method (VT) and fixed thresholding method (FT).
Methods Thirty patients (M:F=20:10, age 51±13 years old) with glioma (n=9,11,10 for WHO grade II,III,IV, respectively) underwent MET PET. In AR, the voxel with the highest intensity in a tumor was chosen as the seed. A neighbor voxel with higher intensity than [mean of current region]*[threshold(%)] was appended to the region. Mean of current region was updated whenever a new voxel joined. An optimal threshold was determined by the sharp point of threshold-volume curve. Two physicians independently measured tumor volume using AR, VT and FT. Contouring was considered as “failed” when the region contained only small parts of the tumor or substantially large parts of normal tissue.
Results All the tumors showed higher MET uptake than background with tumor-to-normal ratio ranging from 1.45 to 5.53. AR succeeded to delineate tumor boundary in 25/30 (83%) cases. Unsuccessful 5 cases consisted of 2 underestimations (grade III and IV) and 3 overestimations (grade II, III, and IV). In terms of successful rate, AR was inferior to VT (29/30, 97%) and superior to FT (15/30, 50%). However, semi-automatic AR reproduced tumor volume between two operators (R2 = 1.00), while VT (R2=0.95) and FT (R2=0.93) showed interoperator variations.
Conclusions A semi-automatic AR is feasible to evaluate glioma expansion using MET PET with the highest reproducibility among the 3 methods investigated