Arterial Spin Labeling Perfusion in Determining the IDH1 Status and Ki-67 Index in Brain Gliomas

Abstract

The aim of the study was to evaluate the relationship between tumor blood flow (TBF) measured by the pseudo-continuous arterial spin labeling (PCASL) method and IDH1 mutation status of gliomas as well as Ki-67 proliferative index. Methods. The study included 116 patients with newly diagnosed gliomas of various grades. They received no chemotherapy or radiotherapy before MRI. IDH1 status assessment was performed after tumor removal in 106 cases—48 patients were diagnosed with wildtype gliomas (Grade 1–2—6 patients, Grade 3–4—42 patients) and 58 patients were diagnosed with mutant forms of gliomas (Grade 1–2—28 patients, Grade 3–4—30 patients). In 64 cases out of 116 Ki-67 index was measured. Absolute and normalized tumor blood flow values were measured on 3D PCASL maps. Results. TBF and normalized TBF (nTBF) in wildtype gliomas were significantly higher than in IDH1-mutant gliomas (p < 0.001). ASL perfusion showed high values of sensitivity and specificity in the differential diagnosis of gliomas with distinct IDH1 status (for TBF: specificity 75%, sensitivity 77.6%, AUC 0.783, cutoff 80.57 mL/100 g/min, for nTBF: specificity 77.1%, sensitivity 79.3%, AUC 0.791, cutoff 4.7). TBF and nTBF in wildtype high-grade gliomas (HGG) were significantly higher than in mutant forms (p < 0.001). ASL perfusion showed the following values of sensitivity and specificity in the diagnosis of mutant HGG and wildtype HGG (for TBF: specificity 83.3%, sensitivity 60%, AUC 0.719, cutoff 84.18 mL/100 g/min, for nTBF: specificity 88.1%, sensitivity 60%, AUC 0.729, cutoff 4.7). There was a significant positive correlation between tumor blood flow and Ki-67 (for TBF Rs = 0.63, for nTBF Rs = 0.61). Conclusion. ASL perfusion may be an informative factor in determining the IDH1 status in brain gliomas preoperative and tumor proliferative activity.

Perifocal Zone of Brain Gliomas: Application of Diffusion Kurtosis and Perfusion MRI Values for Tumor Invasion Border Determination

“…CBF-based detection of the most malignant regions was not performed either. On the contrary, our study included such ROIs in highly perfused enhancing tumor cores as such tumor parts were proven to be the most malignant [45,47].…”

“…This generally supports our previous studies. We consider that biopsy should be taken from the highest perfusion focus defined by ASL (pCASL) perfusion [9,45,47].…”

“…ROI1-the highest perfusion focus in enhancing tumor part according to the fusion of perfusion maps and post-contrast T1 FSPGR. As CBF measured by ASL-perfusion has a correlation with tumor grade and the Ki-67 labeling index (LI), this site was regarded as the most malignant [45]; • ROI2-perifocal infiltrative edema zone according to the fusion of T2-FLAIR CUBE and post-contrast T1 FSPGR (T2-FLAIR CUBE hyperintense signal without signs of pathological contrast enhancement); • ROI3-normal-appearing (intact on conventional MRI) white matter along surgical approach according to T2-FLAIR CUBE images (no pathological changes of MR signal); • ROI4-contralateral (intact on conventional MRI) unaffected hemisphere normalappearing white matter (centrum semiovale).…”

Effects of the Training Data Condition on Arterial Spin Labeling Parameter Estimation Using a Simulation-Based Supervised Deep Neural Network

“…Therefore, only 1 optimized network trained by a sufficiently wide CBF range and the widest ATT range theoretically measurable by the acquisition design is required in clinical practice. However, the effects of extremely high blood flow, which have been reported in some brain tumors, 32 on DNN performance should also be investigated to improve clinical utility.…”

“…setting (ie, data set 15: 0-120 mL/100 g/min for CBF and 0-4500 milliseconds for ATT) was the most accurate because it had the lowest NMAE and NRMSE. Although the CV Net of CBF fluctuated across training data sets, the ATT was generally low, except for the ATT range of 3000 to 4500 milliseconds (data sets [31][32][33][34][35][36]. Based on the final ranking, the best and worst cases were data sets 15 and 36, respectively (Fig.…”