Abstract
BACKGROUND: Glioblastoma is the most aggressive and rapidly growing type of central nervous system tumor. Despite advancements in imaging, no objective measurement for predicting the true extent of glioblastoma has been established. Compared with contrast-enhanced magnetic resonance imaging (MRI), fluid-attenuated inversion recovery (FLAIR) imaging is believed to be more sensitive for detecting infiltrated tumor cells. This study investigates the sensitivity and specificity of preoperative FLAIR imaging to predict glioblastoma true boundaries. METHODS: Our study was retrospectively registered enrolling 20 high-grade glioma patients whose data from 16 patients were analyzed. For each patient, the primary tumor mask was identified on the preoperative FLAIR image covering the whole hyperintense region. Tumor cells infiltration mask was defined on follow-up MRI representing where the tumor recurred. According to automated anatomical labeling 3 (AAL3) and Johns Hopkins University, international consortium of brain mapping, diffusion tensor imaging-white matter-81 labels (JHU ICBM-DTI-81) standard atlases, standard brain was divided into cortical and subcortical regions. Sensitivity and specificity were determined counting the number of brain areas overlapped by the preoperative FLAIR tumor mask and the recurrence tumor mask. RESULTS: The overall sensitivity and specificity was 82.6%, and 84.7%, respectively. Individually, hyperintensity on FLAIR images demonstrated high sensitivity but low specificity in some cases, while in others, the opposite pattern was observed. To validate the reliability of our method, predictive values were defined. The group average positive predictive value and negative predictive value were 50% and 96%, respectively. CONCLUSION: Although FLAIR imaging demonstrates potential in delineating the extent of glioblastoma, its predictive value remains unclear, emphasizing the need for supplementary methodologies to enhance tumor delineation and improve clinical outcomes. CLINICAL TRIAL NUMBER: Not applicable.