MR imaging based fractal analysis for differentiating primary CNS lymphoma and glioblastoma.

OBJECTIVES: The aim of this study was to differentiate primary central nervous system lymphoma (PCNSL) from glioblastomas (GBM) using the fractal analysis of conventional MRI data.

MATERIALS AND METHODS: Sixty patients with PCNSL and 107 patients with GBM with MRI data available were enrolled. Fractal dimension (FD) and lacunarity values of the tumour region were calculated using fractal analysis. A predictive model combining fractal parameters and anatomical characteristics was built using logistic regression. The role of FD, lacunarity and the predictive model in differential diagnosis was evaluated using receiver-operating characteristic (ROC) curve analysis. The association between fractal parameters and anatomical characteristics of tumours was also investigated.

RESULTS: PCNSL had lower FD values (p < 0.001) and higher lacunarity values (p < 0.001) than GBM. ROC curve analysis revealed that FD, lacunarity, and the predictive model could distinguish PCNSL from GBM (area under the curve: 0.895, 0.776, and 0.969, respectively). The following associations were observed between fractal parameters and anatomical characteristics: multiple lesions were significantly associated with higher lacunarity (p = 0.024), necrosis with higher FD (p = 0.027), corpus callosum involvement with higher lacunarity (p < 0.001) in PCNSL and subventricular zone involvement with higher FD (p < 0.001) in GBM.

CONCLUSIONS: The findings of the study indicate that fractal analysis on conventional MRI performs well in distinguishing PCNSL from GBM.

KEY POINTS: • Fractal dimension and lacunarity were capable of differentiating PCNSL from GBM. • PCNSL and GBM exhibited different anatomical characteristics. • Fractal parameters were associated with some of these anatomical characteristics.