Background Glioblastoma (GBM) and primary central nervous system lymphoma (PCNSL) are two distinct malignant brain tumors, and precise preoperative differentiation is crucial for guiding optimal treatments. This study aimed to evaluate the diagnostic value of time-dependent diffusion MRI (td-dMRI)-derived microstructural parameters in differentiating PCNSL from GBM and to correlate these parameters with histopathologic findings. Methods This study included 32 GBM and 19 PCNSL patients who underwent 3.0-T MRI with oscillating gradient spin-echo (OGSE) and pulsed gradient spin-echo (PGSE) sequences. Microstructural parameters [intracellular volume fraction (Vin), cell diameter, cellularity, extracellular diffusivity (Dex)] were compared between the two groups. The area under the receiver operating characteristic curve (AUC) was used to evaluate the diagnostic performance of these parameters. The DeLong test was applied to compare AUC values across parameters. Ridge regression was applied for variable selection, followed by logistic regression to construct a combined diagnostic model. Histopathological validation was performed by correlating td-dMRI parameters with hematoxylin-eosin (H&E)–stained sections. Results In enhancing tumor regions, PCNSL exhibited significantly smaller cell diameter and lower Dex, but higher Vin and cellularity than GBM (all p < 0.001). Vin yielded the highest diagnostic accuracy (AUC = 0.901; sensitivity = 0.737; specificity = 0.906), which was significantly higher than diameter and Dex (all p < 0.05). No significant differences were observed in AUC values between Vin and other ADC-derived parameters, such as ADC0 Hz (AUC = 0.864; sensitivity = 0.895; specificity = 0.781). Ridge regression identified Vin and cellularity as potential independent predictors. However, no statistically robust combined diagnostic model could be established owing to insignificant combined regression results. No significant differences were observed in peritumoral regions. Additionally, Vin showed a strong positive correlation with histopathological nuclei fraction (r = 0.76; p < 0.001). Conclusion Among all td-dMRI-derived microstructural parameters, Vin achieves the highest AUC and serves as a promising biomarker for the preoperative differential diagnosis between GBM and PCNSL.

Microstructure mapping with time-dependent diffusion MRI differentiates primary central nervous system lymphoma from glioblastoma

期刊论文
出版模式：

开放获取
发表日期：

2026年6月1日

查看全文

Jun Wu,
Jue Lu,
Xinli Zhang,
Xiaotong Guo,
Qian Qin,
Jiaqi Chen,
Ning Zheng,
Peng Sun,
Xuan Gao,
Jing Wang

摘要

Background

Glioblastoma (GBM) and primary central nervous system lymphoma (PCNSL) are two distinct malignant brain tumors, and precise preoperative differentiation is crucial for guiding optimal treatments. This study aimed to evaluate the diagnostic value of time-dependent diffusion MRI (t_d-dMRI)-derived microstructural parameters in differentiating PCNSL from GBM and to correlate these parameters with histopathologic findings.

Methods

This study included 32 GBM and 19 PCNSL patients who underwent 3.0-T MRI with oscillating gradient spin-echo (OGSE) and pulsed gradient spin-echo (PGSE) sequences. Microstructural parameters [intracellular volume fraction (V_in), cell diameter, cellularity, extracellular diffusivity (D_ex)] were compared between the two groups. The area under the receiver operating characteristic curve (AUC) was used to evaluate the diagnostic performance of these parameters. The DeLong test was applied to compare AUC values across parameters. Ridge regression was applied for variable selection, followed by logistic regression to construct a combined diagnostic model. Histopathological validation was performed by correlating t_d-dMRI parameters with hematoxylin-eosin (H&E)–stained sections.

Results

In enhancing tumor regions, PCNSL exhibited significantly smaller cell diameter and lower D_ex, but higher V_in and cellularity than GBM (all p < 0.001). V_in yielded the highest diagnostic accuracy (AUC = 0.901; sensitivity = 0.737; specificity = 0.906), which was significantly higher than diameter and D_ex (all p < 0.05). No significant differences were observed in AUC values between V_in and other ADC-derived parameters, such as ADC_0 Hz (AUC = 0.864; sensitivity = 0.895; specificity = 0.781). Ridge regression identified V_in and cellularity as potential independent predictors. However, no statistically robust combined diagnostic model could be established owing to insignificant combined regression results. No significant differences were observed in peritumoral regions. Additionally, V_in showed a strong positive correlation with histopathological nuclei fraction (r = 0.76; p < 0.001).

Conclusion

Among all t_d-dMRI-derived microstructural parameters, V_in achieves the highest AUC and serves as a promising biomarker for the preoperative differential diagnosis between GBM and PCNSL.