Multiparametric and multiregional diffusion features help predict molecule information, grade and survival in lower-grade gliomas: a feasibility study.

Abstract

This study was to investigate the relationship of diffusion features with molecule information, and then predict grade and survival in lower-grade gliomas. 65 patients with primary lower-grade gliomas (WHO Grade II & III) who underwent conventional MRI and diffusion tensor imaging were retrospectively studied. The tumor region was automatically segmented into contrast-enhancing tumor, non-enhancing tumor, edematous and necrotic volumes. Diffusion features, including fractional anisotropy (FA), axial diffusivity, radial diffusivity and apparent diffusion coefficient (ADC), were extracted from each volume using histogram analysis. To estimate molecule biomarkers and predict clinical characteristics of grade and survival, support vector machine, generalized linear model, logistic regression and Cox regression were performed on the related features. The diffusion features in non-enhancing tumor volume showed differences between isocitrate dehydrogenase mutant and wild-type gliomas. And the mean accuracy of support vector machine classifiers was 0.79. Ki-67 labeling index was correlated with these features, which were combined to significantly estimate Ki-67 expression level (r = 0.657, p < 0.001). These features also showed differences between Grade II and III gliomas. A combination of them for grade classification resulted in an area under the curve of 0.914 (0.857-0.971). Mean FA and fifth percentile of ADC were independently associated with overall survival, with lower FA and higher ADC showing better survival outcome. In lower-grade gliomas, multiparametric and multiregional diffusion features could help predict molecule information, histological grade and survival. The multi parametric diffusion features in non-enhancing tumor were associated with molecule information, grade and survival in lower-grade gliomas.