Assessment of pTERT Subtypes of Glioblastoma by Quantitative FLAIR Analysis Based on Fusing MRI Images

Mutations in the telomerase reverse transcriptase promoter (pTERT) are important molecular markers in glioblastoma (GBM). Although several imaging-based approaches have attempted to predict pTERT mutation status preoperatively, the value of quantitative metrics extracted from lesion subregions remains unclear. This study investigated whether quantitative FLAIR metrics derived from contrast-enhanced T1-weighted imaging (CE T1WI)-FLAIR fused images contribute to the differentiation of pTERT subtypes in GBM.

MRI and clinical data from 135 GBM patients, 94 with pTERT-mutant (pTERTm) and 41 with pTERT-wild-type (pTERTw) tumors, were retrospectively analyzed. Patients were randomly assigned to training and validation cohorts in a 7:3 ratio. Clinical characteristics and conventional MRI variables were compared between pTERTm and pTERTw groups. FLAIR signal intensity (SI) metrics were measured in three subregions on CE T1WI-FLAIR fused images: the enhancement region, the edema region (non-enhancing), and the whole lesion (enhancement + edema). Significant variables identified by logistic regression were incorporated into clinical, MRI, and combined predictive models. Model performance was internally evaluated using leave-one-out cross-validation (LOOCV) in the training cohort and externally assessed in the validation cohort.

Significant differences between pTERTm and pTERTw groups were observed in age, FLAIR SI standard deviation (FLAIR_SD) and relative FLAIR SI (rFLAIR) of the edema region, and FLAIR_SD of the enhancement region (all p < 0.05). Logistic regression identified older age (> 42.5 years; OR = 1.09; p = 0.002), higher FLAIR_SD in the enhancement region (> 62.45; OR = 1.01; p = 0.027), and higher rFLAIR in the edema region (> 1.706; OR = 7.49; p = 0.025) as independent predictors of pTERTm. In the training cohort, the combined model achieved an area under the ROC curve (AUC) of 0.833, outperforming the clinical model (0.675) and MRI-based model (enhancement-region FLAIR_SD: 0.737; edema-region rFLAIR: 0.699). The combined model achieved the highest predictive performance, with LOOCV in the training cohort yielding a mean AUC of 0.784 (95% CI: 0.672-0.895) and external validation showing an AUC of 0.667 (95% CI: 0.466-0.867).

Quantitative FLAIR metrics extracted from subregions on CE T1WI-FLAIR fused images differ significantly between pTERTm and pTERTw GBM. Subregional quantitative analysis may therefore contribute to noninvasive preoperative prediction of pTERT mutation status.