Predicting cancer malignancy and proliferation in glioma patients: intra-subject inter-metabolite correlation analyses using MRI and MRSI contrast scans.

BACKGROUND

The non-invasive characterization of glioma metabolites would greatly assist the management of glioma patients in the clinical setting. This study investigated the applicability of intra-subject inter-metabolite correlation analyses for differentiating glioma malignancy and proliferation.

METHODS

A total of 17 negative controls (NCs), 39 low-grade gliomas (LGGs) patients, and 25 high-grade gliomas (HGGs) subjects were included in this retrospective study. Amide proton transfer (APT) and magnetization transfer contrast (MTC) imaging contrasts, as well as total choline/total creatine (tCho/tCr) and total N-acetylaspartate/total creatine (tNAA/tCr) ratios quantified from magnetic resonance spectroscopic imaging (MRSI) were co-registered voxel-wise and used to produce three intra-subject inter-metabolite correlation coefficients (IMCCs), namely, R , R , and R . The correlation between the IMCCs and tumor grade and Ki-67 labeling index (LI) for tumor proliferation were explored. The differences in the IMCCs between the three groups were compared with one-way analysis of variance (ANOVA). Finally, regression analysis was used to build a combined model with multiple IMCCs to improve the diagnostic performance for tumor grades based on receiver operator characteristic curves.

RESULTS

Compared with the NCs, gliomas showed stronger inter-metabolic correlations. R was significantly different among the three groups (NC . LGGs . HGGs: -0.18±0.38 . -0.40±0.34 . -0.70±0.29, P<0.0001). No significant differences were detected in R among the three groups. R and R correlated significantly with tumor grade (R=-0.41, P=0.001 and R=0.448, P<0.001, respectively). However, only R was mildly correlated with Ki-67 (R=-0.33, P=0.02). R and R achieved areas under the curve (AUCs) of 0.754 and 0.71, respectively, for differentiating NCs from gliomas; and 0.77 and 0.78, respectively, for differentiating LGGs from HGGs. The combined multi-IMCCs model improved the correlation with the Ki-67 LI (R=0.46, P=0.0008) and the tumor-grade stratification with AUC increased to 0.85 (sensitivity: 80.0%, specificity: 79.5%).

CONCLUSIONS

This study demonstrated that glioma patients showed stronger inter-metabolite correlations than control subjects, and the IMCCs were significantly correlated with glioma grade and proliferation. The multi-IMCCs combined model further improved the performance of clinical diagnosis.