OBJECTIVES

The composition and extent of edema (ED) region can reflect the aggressive degree of glioblastoma (GBM). Investigating its heterogeneity pattern and identifying the high-risk habitat may provide important prognostic information.

METHODS

We prospectively collected 122 GBM patients from the Cancer Imaging Archive (TCIA) and 65 GBM patients from the local institution for the training cohort and external test cohort respectively. The signal intensities of ED region from T1-weighted contrast-enhanced images (T1CE) and T2-weighted fluid-attenuated inversion recovery images (FLAIR) were pooled together from each patient as a global matrix. Then, K-means clustering was applied, which could segment ED regions into several habitats (i.e., subregions). A group of radiomics features were extracted and radiomics signatures (RadScores) were derived. The high-risk habitat was identified and evaluated in light of the prognostic at the intra-regional, inter-regional, and model levels. Molecular analysis was also conducted to investigate the potential of the high-risk habitat in complementing biological information.

RESULTS

After three levels comparison, the high-risk habitat was determined. When combing with the RadScores of enhanced tumor (ET), the concordance index (C-index) was leveraged from 0.658 to 0.677. When combining with clinical factors and RadScores of ET, the C-index increased to 0.770. For molecular analysis, we observed a more significant difference among groups in survival prediction after uniting MGMT methylation status and the high-risk habitat signature.

CONCLUSIONS

This study demonstrates that investigating the spatial heterogeneity of ED and identifying the high-risk habitat within it may provide more references for GBM treatment and prognosis studies.