What type of study is this?

August 20, 2025Open Access

Tumor Habitats Based on Multiparametric MRI Distinguish Atypical Glioblastoma From Primary Central Nervous System Lymphoma: Imaging‐Pathologic Correlation

Key Points

The model effectively distinguishes atypical glioblastoma from PCNSL using MRI features, enhancing diagnostic accuracy.
Statistically significant differences in voxel fractions of tumor habitats were observed between atypical glioblastoma and PCNSL groups.
A logistic regression model incorporating habitat types, edema index, and apparent diffusion coefficient was validated with an AUC of 0.851.
Higher levels of vasculogenic mimicry in the PCNSL group reveal underlying pathological differences between these tumors.

Abstract

ABSTRACT Background Atypical glioblastoma (GBM) (minimal or no necrosis on MRI) and primary central nervous system lymphoma (PCNSL) are difficult to distinguish on MRI; whether tumor habitat can more accurately distinguish atypical GBM from PCNSL remains uncertain. Purpose To evaluate the diagnostic performance with tumor habitats, apparent diffusion coefficient (ADC), and edema index (EI) to distinguish atypical GBM from PCNSL. Study Type Retrospective. Population One hundred twenty‐five patients (63 male and 62 female) diagnosed with atypical GBM or PCNSL were included. Field Strength/Sequence 1.5 T and 3.0 T, Axial ADC and T1 contrast‐enhanced spin‐echo inversion recovery sequence (T1‐CE). Assessment The tumor habitat was derived using T1‐CE and ADC sequences. Based on this tumor habitat, EI and relative ADC (rADC), we constructed a model. Statistical Tests Logistic regression; Akaike Information Criterion; Receiver operating characteristic (ROC) curves, calibration curves, and Decision Curve Analysis. Results Three tumor habitats were identified: high‐enhancement cellular habitat (Habitat 1), low‐enhancement cellular habitat (Habitat 2), and nonviable tissue habitat (Habitat 3). The voxel fraction of the three tumor habitats in atypical GBM and PCNSL groups shows statistically significant differences. The EI of patients in the PCNSL group was significantly higher than that of the patients in atypical GBM . A model was established incorporating the parameters Habitat 2, Habitat 3, EI , and rADCmean . The model exhibits excellent discriminative ability in the training set ( AUC = 0.851, 95% CI : 0.781–0.921) and validation set ( AUC = 0.807, 95% CI : 0.724–0.889). Histopathological evaluation showed that vasculogenic mimicry ( VM ) levels were significantly higher in the PCNSL group. Multiple linear regression analysis showed a significant correlation between habitat voxel fraction and VM levels. Data Conclusion A model built based on tumor habitat, EI, and rADCmean can differentiate atypical GBM from PCNSL preoperatively. The differences in VM levels are one of the pathological mechanisms underlying the variations in tumor habitats between atypical GBM and PCNSL. Level of Evidence 5. Technical Efficacy Stage 3.

Read Full Paperexternally

KI fragen

Bookmark

View Full Paper