Exhibit Hall | Forum 2
Purpose: Gamma knife (GK) radiosurgery is typically planned on MR images using water-equivalent dose calculation methods. The aim of this study is to evaluate MR-generated synthetic CT (synCT) for GK dose calculation with tissue inhomogeneity correction.
Methods: The university-industry collaboration study consists of developing a deep-learning (DL) method to generate synCT from routine simulation MR images, and assessing the potential of synCT for GK dose calculation. Treatment planning T1-weighted MRI and CT images of fifty patients who received GK were provided to train a synCT generation algorithm using cycle generative adversarial network (cycleGAN) with mutual information constraints. synCT of 10 testing GK patients were generated from the MR and compared with corresponding CT. Using GammaPlan 11.3.1, the GK plans created on MRI using the TMR10 method were re-calculated using the convolution calculation algorithm with electron density maps from the CT and synCT. The doses were evaluated with CT-calculated doses by dose-volume histogram metrics on 20 brain tumor targets of the testing cases.
Results: The mean absolute error (MAE ± std) between synCT and CT were 118 ± 30 HU within MR-delineated skulls for the testing cases, 48 ± 15 HU for the soft tissue and 297 ± 87 HU for the bone. Percentage differences in GTV D95% and V12Gy from CT-calculated doses were 3.6 ± 4.5%, 0.2 ± 1.2%, respectively, for synCT dose calculation, and became 9.3 ± 3.1%, 8.9 ± 1.1% when TMR10 was used.
Conclusion: synCT and CT yield equivalent dose calculation for GK Plans, while the water-based TMR method shows appreciable dose discrepancies. MR-generated synCT may enable tissue inhomogeneity-corrected dose calculation for current MR-based GK Planning.
Funding Support, Disclosures, and Conflict of Interest: This research is supported by Elekta AB
Not Applicable / None Entered.