MRI-based quantification of intratumoral heterogeneity for differentiating glioblastoma from solitary brain metastasis: a two-center study.

OBJECTIVES

To evaluate the value of MRI-based quantification of Intratumoral Heterogeneity (ITH) for differentiating Glioblastoma (GBM) from Solitary Brain Metastasis (SBM).

METHODS

This retrospective study included contrast-enhanced T1-weighted imaging (CE-T1WI) data from 355 patients diagnosed with GBM or SBM at two institutions. Data from Institution 1 (n = 267) were used for predictive model development and as an internal test dataset, while data from Institution 2 (n = 88) constitute the external test dataset. To quantify ITH, 2.5D data were utilized to calculate the ITH-scores by integrating local radiomics features and global pixel distribution patterns, and the ITH model was constructed on this basis. Moreover, a radiomics model and a combined radiomics-ITH model were established. The predictive performance of three models was comprehensively evaluated using calibration curves, decision curve analysis (DCA), accuracy, and AUC.

RESULTS

The radiomics model achieved an AUC of 0.64 in internal test dataset and 0.62 in external test dataset, whereas the ITH model yielded an AUC of 0.82 in internal test dataset and 0.81 in external test dataset, the ITH model exhibited significantly better predictive performance than the radiomics model. The predictive efficacy of the combined radiomics-ITH model was further improved, with an AUC of 0.90 in internal test dataset and 0.87 in external test dataset. Calibration curves and DCA confirmed the combined model possessed favorable calibration ability and good clinical applicability.

CONCLUSIONS

MRI-based quantification of ITH may serve as an effective noninvasive tool for differentiating GBM from SBM and assist in clinical decision-making.