A Comparison of Modalities to Differentiate Radiation Necrosis from Tumor Progression: A Diagnostic Meta-Analysis

Abstract

<h3>Purpose/Objective(s)</h3> Cerebral radiation necrosis is a complication following radiation treatment to the brain. It is hypothesized that radiation treatment causes vascular injury, leading to tissue ischemia and then radiation necrosis. Radiation necrosis is a delayed phenomenon, occurring several months to years after the completion of radiation therapy. Radiation necrosis can be difficult to diagnose as it appears similar to tumor progression on standard magnetic resonance imaging (MRI). The ability to distinguish the two is important for further treatment strategies, and there is no current guideline on which imaging modality should be used. Various imaging modalities have been studied as an alternative to MRI, including, but not limited to, MR perfusion techniques (MRP), MR spectroscopy (MRS), single photon emission computed tomography (SPECT), and positron emission tomography (PET). We aim to review the existing literature and perform a diagnostic meta-analysis comparing these different imaging modalities to differentiate radiation necrosis from tumor progression. <h3>Materials/Methods</h3> A systematic review adherent to PRISMA guidelines was performed using Scopus and PubMed databases. Sensitivities and specificities were pooled using the inverse-variance method, and 95% confidence intervals were calculated. To compare modalities, Euclidean distances are calculated as the absolute distance from the point (1.0, 1.0) of a theoretical ideal classifier on a graph of sensitivity vs. specificity. For evaluation of Euclidean distance, the lower the value, the closer the test is to optimal, which would be a sensitivity and specificity of 1. <h3>Results</h3> The analysis includes 113 studies with a total of 200 imaging data points, including the following imaging modalities: MRI (n = 13), MRS (n = 21), MRP (n = 12), Dynamic Contrast Enhanced (DCE) MRI (n = 4), Dynamic Susceptibility Contrast MRI (n = 20), MR arterial spin labeling (ASL) (n = 3), Diffusion Weighted Imaging (n = 11), Diffusion Tensor Imaging (DTI) (n = 2), PET (n = 88), and SPECT (n = 26). MRS had the highest pooled sensitivity at 0.92 (95% CI: 0.84 - 0.96), followed by DCE MRI (0.88; 95% CI: 0.70 - 0.96) and then SPECT (0.88; 95% CI: 0.82 - 0.93). The highest pooled specificities were DTI at 0.90 (95% CI: 0.69 - 0.98), SPECT at 0.88 (95% CI: 0.83 - 0.92), and DCE MRI at 0.85 (95% CI: 0.71 - 0.93). For Euclidean distances, SPECT had the lowest at 0.163 (95% CI: 0.106 - 0.248), followed by MRS (0.167; 95% CI: 0.100 - 0.275), and DCE MRI (0.185; 95% CI: 0.077 - 0.414). MRI had the highest Euclidean distance at 0.373 (95% CI: 0.238 - 0.542). <h3>Conclusion</h3> The results from this diagnostic meta-analysis evaluating pooled sensitivities, pooled specificities, and Euclidean distances indicate that SPECT and MRS appear to be promising imaging modalities for the differentiation of radiation necrosis from tumor progression. This study may have implications in the formation of imaging guidelines for this purpose.