Planning dosimetry for ⁹⁰Y radioembolization with glass microspheres: Evaluating the fidelity of ^99mTc-MAA and partition model predictions

Purpose: ^99mTc-MAA-SPECT/CT may be used in ⁹⁰Y-glass microsphere radioembolization treatment planning to assess perfused liver volumes and absorbed dose distributions. The partition model (PM) offers a more detailed planning dosimetry option beyond the single-compartment model more traditionally used in ⁹⁰Y radioembolization. As ⁹⁰Y radioembolization treatments shift toward activities and doses that aim to achieve tumor control, accurate and reliable treatment planning dosimetry for both tumors and normal liver (NL) becomes more critical. In this work, we explore the accuracy and precision of ⁹⁰Y dosimetry predictions from pretherapy ^99mTc-MAA and PM. Methods: Both PM and voxel dosimetry models were used to calculate tumor and NL mean doses using both planning ^99mTc-MAA and verification ⁹⁰Y-SPECT/CT in this retrospective analysis of hepatocellular carcinoma cases treated with glass microspheres (NCT01900002, n = 32). Linear regression models were developed at first access, and then later correct, the estimates by (a) ^99mTc-MAA for ⁹⁰Y voxel dosimetry and (b) ^99mTc-MAA PM for voxel dosimetry, separately for both tumors and NL. Bland-Altman analysis was then used to evaluate the accuracy and precision of the regression model predictions with the mean bias and 95% prediction intervals (PI, ±1.96σ). Two categories of cases were stratified (catheter matched vs catheter unmatched) by establishing the level of ^99mTc-MAA and ⁹⁰Y catheter position alignment. Only catheter-matched cases were included in the ^99mTc-MAA vs ⁹⁰Y voxel dosimetry comparison, while all cases were used to compare dosimetry models (PM vs voxel). Results: Half (16/32) of cases were deemed catheter matched. ^99mTc-MAA could reliably predict NL doses in catheter-matched cases after application of the linear model, with mean bias (PI) of −1% (±31%). PM was equivalent to voxel dosimetry for NL doses with mean bias (PI) of 0% (±1%). Even among catheter-matched cases, ^99mTc-MAA planning for ⁹⁰Y tumor voxel doses was poor, overestimating dose by an average of nearly 40%. Upon application of the linear model, ^99mTc-MAA predictions for ⁹⁰Y tumor voxel dose were only minimally biased (−4%) but possessed very large PI (±104%). PM predictions for tumor voxel dose using the linear model also showed small bias (−6%) but maintained similarly high PI of ±90%. Cases with tumors representing a large majority (>80%) of the total tumor volume demonstrated the best scenarios for ^99mTc-MAA and PM tumor dose predictions, with mean biases (PI) of −3% (±53%) and −4% (±21%), respectively. Conclusion: The unconditional use of ^99mTc-MAA to predict ⁹⁰Y dosimetry across all cases is not recommended due to: (a) demonstrated the risk of unmatched catheter positions between procedures, and (b) large bias and uncertainty in ^99mTc-MAA predictions in cases with matched catheter locations. However, NL voxel dose predictions with ^99mTc-MAA are clinically viable and either PM or voxel dosimetry can be used to produce equivalent predictions. Both ^99mTc-MAA and PM can provide tumor dose predictions with potential clinical utility, but only in catheter-matched cases and with tumors comprising a clear majority (>80%) of the total tumor volume. These findings stratify the predictive fidelity of ^99mTc-MAA- and PM-based treatment planning for ⁹⁰Y dosimetry in improving treatment outcomes.