Shu Zhang1, Jie Wang1, Kai Wang1, Xiaotong Li1, Xiaobin Zhao1, Qian Chen1, Wei Zhang1, Lin Ai2. 1. Department of Nuclear Medicine, Beijing Tiantan Hospital, Capital Medical School, No. 119, West Road of South 4th Ring, Beijing, China. 2. Department of Nuclear Medicine, Beijing Tiantan Hospital, Capital Medical School, No. 119, West Road of South 4th Ring, Beijing, China. Electronic address: ailin@bjtth.org.
Abstract
OBJECTIVES: To investigate the value of the 18F-FDG PET/MRI multiparametric model in the differentiation of high-grade glioma (HGG) and primary central nervous system lymphoma (PCNSL), with emphasis on the quantitative analysis of the enhancing tumor (ET) and non-enhancing peritumoral region (PTR). METHODS: Forty-five patients with HGG and 20 patients with PCNSL who underwent simultaneous 18F-FDG PET, arterial spin labelling perfusion-weighted imaging and diffusion-weighted imaging with hybrid PET/MRI before treatment were retrospectively enrolled. The relative maximum standardized uptake value (rSUVmax), relative maximum cerebral blood flow (rCBFmax) and relative minimum apparent diffusion coefficient (rADCmin) in both the ET and NPR were calculated and compared between HGG and PCNSL. Multivariate logistic regression was used to determine the best logistic regression model (LRM) for classification. Receiver operating curve analysis was used to assess diagnostic performance. RESULTS: In the ET, HGG showed significantly lower rSUVmax values but higher rCBFmax and rADCmin than PCNSL (all P < 0.05). In the PTR, HGG demonstrated significantly higher rSUVmax and rCBFmax but lower rADCmin than PCNSL (all P < 0.05). Multivariate logistic regression based on quantitative parameters revealed that the LRM consisting of rSUVmax_ET, rADCmin_ET and rCBFmax_PTR had significantly improved diagnostic performance in differentiating HGG from PCNSL than single parameter alone, with an AUC of 0.980 and an accuracy of 95.4%. Multivariate logistic regression incorporating quantitative parameters and conventional MRI features revealed that the LRM consisting of rSUVmax_ET, rCBFmax_PTR and enhancement pattern yielded a slightly higher AUC of 0.989 and an identical accuracy of 95.4%. No significant difference in AUCs was detected between the two LRMs (P = 0.233). CONCLUSIONS: Multiparametric 18F-FDG PET/MRI diagnostic model based on conventional MRI features and quantitative analysis of the enhancing tumors and peritumoral regions is superior to single parameter in the differentiation of HGG and PCNSL, which should be considered in the clinical practice.
OBJECTIVES: To investigate the value of the 18F-FDG PET/MRI multiparametric model in the differentiation of high-grade glioma (HGG) and primary central nervous system lymphoma (PCNSL), with emphasis on the quantitative analysis of the enhancing tumor (ET) and non-enhancing peritumoral region (PTR). METHODS: Forty-five patients with HGG and 20 patients with PCNSL who underwent simultaneous 18F-FDG PET, arterial spin labelling perfusion-weighted imaging and diffusion-weighted imaging with hybrid PET/MRI before treatment were retrospectively enrolled. The relative maximum standardized uptake value (rSUVmax), relative maximum cerebral blood flow (rCBFmax) and relative minimum apparent diffusion coefficient (rADCmin) in both the ET and NPR were calculated and compared between HGG and PCNSL. Multivariate logistic regression was used to determine the best logistic regression model (LRM) for classification. Receiver operating curve analysis was used to assess diagnostic performance. RESULTS: In the ET, HGG showed significantly lower rSUVmax values but higher rCBFmax and rADCmin than PCNSL (all P < 0.05). In the PTR, HGG demonstrated significantly higher rSUVmax and rCBFmax but lower rADCmin than PCNSL (all P < 0.05). Multivariate logistic regression based on quantitative parameters revealed that the LRM consisting of rSUVmax_ET, rADCmin_ET and rCBFmax_PTR had significantly improved diagnostic performance in differentiating HGG from PCNSL than single parameter alone, with an AUC of 0.980 and an accuracy of 95.4%. Multivariate logistic regression incorporating quantitative parameters and conventional MRI features revealed that the LRM consisting of rSUVmax_ET, rCBFmax_PTR and enhancement pattern yielded a slightly higher AUC of 0.989 and an identical accuracy of 95.4%. No significant difference in AUCs was detected between the two LRMs (P = 0.233). CONCLUSIONS: Multiparametric 18F-FDG PET/MRI diagnostic model based on conventional MRI features and quantitative analysis of the enhancing tumors and peritumoral regions is superior to single parameter in the differentiation of HGG and PCNSL, which should be considered in the clinical practice.