M. Allan Thomas¹*, Benjamin P. Lopez¹, Adam Neff³, Armeen Mahvash², S. Cheenu Kappadath¹, (1) Department of Imaging Physics, UT MD Anderson Cancer Center, Houston, TX, (2) Department of Interventional Radiology, UT MD Anderson Cancer Center, Houston, TX, (3) MIM Software, Inc., Cleveland, OH

(Wednesday, 7/15/2020) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Room: Track 1

Purpose: Accurate determination of ?°Y net administered activity (AA) in ?°Y-radioembolization therapy is critical for dosimetry and assessment of potential mis-administrations. The accuracy and uncertainty in determination of AA depends on dose calibrator configuration and measurement of activity residuals post-therapy. While glass-microspheres typically have high fraction of AA and well-defined procedures for residual assessments, resin-microspheres are more prone to stasis and have poorly defined procedures for residual calculation. Consequently, determination of AA for ?°Y-radioembolization with resin-microspheres is challenging. Here, we propose a model to compute AA from post-therapy ?°Y-SPECT/CT.


Methods: Ground-truth AA for 67 cases of ?°Y-radioembolization using glass-microspheres was determined at time of post-therapy ?°Y-SPECT/CT imaging with adjustments for measured residuals (median=1.0%, 25%-75% range=0.4-2.9%) and ?°Y decay between administration and imaging (152, 120-188 min). Univariate linear models were developed based on leave-one-out cross validation to predict AA using reconstructed SPECT counts from two different VOIs: 1) full-FOV and 2) liver-only. Liver VOIs were generated by an AI-based auto-segmentation algorithm (MIM Software). The mean bias for accuracy and 95% prediction intervals (PI) were assessed with Bland-Altman analysis for both VOIs.


Results: Both SPECT VOIs produced excellent AA predictions relative to ground truth with mean absolute errors of 7.0% (r²=0.98) and 6.4% (r²=0.99) for full-FOV and liver-only VOIs, respectively. With Gaussian distribution of residual errors, mean errors and PI were 0.30% ± 17.7% for full-FOV and -0.50% ± 15.7% for liver-only VOIs. AA predictions based on full-FOV VOIs with PI<18% enable straightforward assessment of potential mis-administrations. AA predictions based on liver-only VOIs additionally provide liver activity with mean absolute errors <6% for dosimetry calculations.


Conclusion: Prediction models based on reconstructed SPECT counts offer an accurate and direct method for determination of AA values following ?°Y-radioembolization. Characterization of AA following ?°Y-radioembolization with resin-microspheres based on these models is underway.

Keywords

SPECT, Image Analysis

Taxonomy

IM- SPECT : Quantitative imaging

Contact Email