Abstract: Background and purpose: Radiomics is a hot topic in recent years, which can quantify tumor heterogeneity, and is widely used in lesion characterization, clinical staging, efficacy evaluation and risk factor stratification. This study was designed to investigate the differential diagnosis of glioblastoma and primary central nervous system lymphoma by conventional magnetic resonance imagine (MRI) texture analysis. Methods: The clinical and imaging data of 35 cases of glioblastoma and 15 cases of primary central nervous system lymphoma confirmed by postoperative pathology from Jun. 2012 to Jul. 2017 in the First Affiliated Hospital of Wannan Medical College were retrospectively analyzed. All patients underwent conventional MRI scan, including axial T1WI, T2WI and T2WI-weighted fluid-attenuated inversion recovery (T2-FLAIR). The texture features of the lesion was extracted by MaZda software to manually draw the region of interest (ROI) on the maximum level of tumor delineation on three plain scan sequences. Results: Through a statistical screening of a large number of texture features, the differences in T1WI autocorrelation, T1WI entropy, T2WI mean, T2-FLAIR mean and T2-FLAIR entropy among the galactic co-occurrence matrix parameters were statistically significant between glioblastoma and primary central nervous system lymphoma. Logistic regression analysis showed the area under the receiver operating characteristic curve (ROC) of the model was 0.94. Conclusion: Conventional MRI texture analysis provides reliable and quantified objective basis without enhanced examination, for the differential diagnosis of glioblastoma and primary central nervous system lymphoma.

Key words: Glioblastoma, Primary central nervous system lymphoma, Magnetic resonance imaging, Texture analysis