Apparent diffusion coefficient histogram analysis for prediction of prognosis in glioblastoma

To investigate the potential to predict prognosis of glioblastoma (GBM) patients by analysis of the broader and lower values in the lower distribution of apparent diffusion coefficient (ADC_L) (B&L-ADC_L) values in the ADC histogram.

Presurgical publicly available diffusion-weighted images (DWI) and contrast-enhanced T1-weighted images from 76 GBM patients were analyzed. With applied 2-mixture normal distribution in the ADC histogram of enhanced lesions on T1-weighted images, the mean and width of ADC_L were analyzed. We dichotomized the lower mean of ADC_L (L-ADC_L) and the broader width of ADC_L (B-ADC_L) at their own average. B&L-ADC_L was defined as B-ADC_L with L-ADC_L. Progression-free survival (PFS) and overall survival (OS) were determined by using Cox proportional hazards analysis and the Kaplan-Meier method with the log-rank test. The difference between PFS and OS was calculated.

Six (7.89%) patients had B&L-ADC_L values. B&L-ADC_L was strongly associated with poor PFS (hazard risk: 5.747; P=0.002) and OS (hazard risk: 3.331; P=0.018). There were significant differences in PFS (median, 77 vs. 302 days; P<0.001) and OS (median, 199 vs. 472 days; P=0.004) between the patents with and without B&L-ADC_L. There was no significant difference in the OS-PFS duration difference between the patients with (median, 79 days) and without B&L-ADC_L (median, 132 days) (P=0.348).

In this study, B&L-ADC_L from pretreatment ADC analysis predicted poor PFS. B&L-ADC_L may indicate higher cellularity and heterogeneity in GBM tumor tissue.