Voxel-based analysis of (201)Tl SPECT for grading and diagnostic accuracy of gliomas: comparison with ROI analysis

PURPOSE: The aim of this retrospective study was to assess the utility of a voxel-based analysis (VBA) method for (201)Tl SPECT in glioma, compared to conventional ROI analysis. METHODS: We recruited 24 patients with glioma (high-grade 15; low-grade 9), for whom pre-operative (201)Tl SPECT and MRI were performed. SPECT images were coregistered with MRI. The uptake ratio (UR) images of tumor to contralateral normal tissue were measured on early and delayed images, and the (201)Tl retention index (RI) map was calculated from the early and delayed uptake ratio maps. In the ROI analysis, tumors were traced on a UR map, and the mean and maximal uptake ratio values on the early images were, respectively, defined as the mean and maximal UR. The mean and maximal RI values (mean and maximal RI) were calculated by division of the mean and maximal UR, respectively, on the delayed image by the mean and maximal UR on the early image. For the RI map calculated voxel by voxel, the maximal RI value was defined as VBA-RI. We evaluated sensitivity and accuracy of differential analysis with the mean and maximal UR, RI, and VBA-RI. RESULTS: The high- and low-grade groups showed no significant difference in mean and maximal RI (0.98 ± 0.12 vs. 1.05 ± 0.09 and 0.98 ± 0.18 vs. 1.05 ± 0.14, respectively). The AUC and accuracy of the mean and maximal RI were 0.681 and 66.7 %, and 0.622 and 62.5 %, respectively. In contrast, VBA-RI was higher in high-grade than in low-grade glioma (1.69 ± 0.27 vs. 0.68 ± 0.66, p < 0.001). The AUC and accuracy of VBA-RI were 0.963 and 95.8 %, which are higher than those obtained for mean (p < 0.05) and maximal RI (p < 0.01). There was no significant difference in ROC between the VBA-RI and the mean UR (0.911, p = 0.456) and maximal UR (0.933, p = 0.639); however, the AUC, sensitivity, and diagnostic accuracy of VBA-RI were all higher than those of the mean and maximal UR. CONCLUSION: The voxel-based analysis method of (201)Tl SPECT may improve diagnostic performance for gliomas, compared with ROI analysis.