Voxel-based dosimetry with integrated Y-90 PET/MRI and prediction of response of primary and metastatic liver tumors to radioembolization with Y-90 glass microspheres.

PURPOSE

In this study, we aimed to evaluate the response of the primary and metastatic liver tumors to radioembolization with Y glass microspheres and investigate its correlations with dosimetric variables calculated with Y PET/MRI.

METHODS

In this ambispective study, 44 patients treated with Y glass microspheres and imaged with Y PET/MRI were included for analysis. Dosimetric analysis was performed for every perfused lesion using dose-volume histograms. Response was assessed by comparing pre-treatment and follow-up total lesion glycolysis (TLG) values derived from F-FDG PET imaging. The relationship between ΔTLG and log-transformed dosimetric variables was analyzed with linear mixed effects regression models. ROC analyses were performed to compare discriminatory power of the variables in predicting response and complete response.

RESULTS

Regression and ROC analyses demonstrated that mean tumor dose and almost all D values were statistically significant predictors of treatment response and complete treatment response. Specifically, D60, D70 and D80 values exhibited significantly higher discriminatory power for predicting treatment response compared to the mean dose (D) delivered to tumor. High specificity cut-off values to predict response were determined as 160.75 Gy for D, 95.50 Gy for D60, 89 Gy for D70, and 59.50 Gy for D80. Similarly, high-specificity cut-off values to predict complete response were 262.75 Gy for D, 173 Gy for D70, 140.5 Gy for D80, and 100 Gy for D90.

CONCLUSION

In this study, we demonstrated that voxel-based dosimetry with post-treatment Y PET/MRI can predict response to treatment. D60, D70 and D80 variables also did have greater discriminatory power compared to D in prediction of response. In addition, we present high-specificity cut-offs to predict response (CR + PR) and complete response (CR) for both D and several D variables derived from dose-volume histograms.