Program Information
Evaluation of Synthetic CTs Generated Using MR-SIM Data
J Kim*, C Glide-Hurst , A Doemer , N Wen , I Chetty , Henry Ford Health System, Detroit, MI
Presentations
TH-A-BRF-3 Thursday 7:30AM - 9:30AM Room: Ballroom FPurpose: To describe and evaluate a novel algorithm for generating synthetic CT images from MR-SIM data for dose calculations in MR-only treatment planning.
Methods: A voxel-based weighted summation method was implemented to generate synthetic CT (synCT) images. MR data were acquired using Philips 1.0T Panorama high-field open MR-SIM. Retrospective patient data from seven prostate patients and one brain patient (three lesions) enrolled in an IRB-approved study were used. 3D T1-weighted fast field echo and 3D T2-weighted turbo spin echo sequences were utilized for all patients. A 3D balanced turbo field echo sequence using spectral presaturation with inversion recovery was acquired for prostate patients, but 3D ultra-short echo time (UTE)-DIXON was instead acquired for the brain patient to amplify bone signal for semi-automatic bone segmentation. Weight optimization was performed using a training subset of patients. HU value differences between planning CT and synCTs were analyzed using mean absolute error (MAE). Original patient CT-based treatment plans were mapped onto synCTs, dose was recalculated using original leaf motion and MU values, and DRRs were generated. Dose-volume metrics and gamma analysis were used for dosimetric evaluation.
Results: Average whole-body MAE of synCTs across all patients was 75+12 HU. In prostate cancer patients, average HU difference between planning and synCTs was 0.9±1.0% for soft tissue structures and 4.3±2.5% for bony structures. DRRs were generated from synCTs and qualitatively showed good geometric agreement with planning CT-generated DRRs. D99, mean dose, and maximum dose to CTV calculated using the synCT remained within 1.2% of planning CT-based dose calculations. All gamma analysis evaluated at 2%/2mm dose difference/distance to agreement) pass rates were greater than 95% with an average of 99.9±0.1% for prostate patients and 98.4±2.2% for three brain lesions.
Conclusion: SynCTs were generated with clinically acceptable accuracy comparable to planning CTs, enabling dose computations for MR-only simulation.
Funding Support, Disclosures, and Conflict of Interest: Research supported in part by a grant from Philips HealthCare (Best, Netherlands).
Contact Email: