Purpose: Gamma knife radiosurgery (GKS) cranial dose calculation using TMR 10 algorithm ignores tissue inhomogeneity which can introduce errors. Electron densities required for tissue inhomogeneity correction can be obtained from MRI based synthetic CT (sCT). The study estimates the dose differences as a result of tissue inhomogeneity with TMR10 and convolution algorithm (CA) on sCT.
Methods: Post contrast T1-weighted (T1p) MRI based sCT were generated from two patients who underwent GKS, using cycle consistent generative adversarial networks. Twelve patients who underwent external beam radiotherapy and had T1p images were selected for training and one patient for testing. The training was evaluated with leave-one-out cross validation. Image intensities between patient CT (pCT) and sCT were analyzed with HU histogram comparison and mean absolute error (MAE). Three plans were designed with single 8mm shots at three different locations – near skull, near air cavities (sinuses) and near corpus callosum (homogeneous tissues). Shot times, indicative of dose differences, for TMR10 calculated on pCT and CA on sCT were compared with CA calculated on pCT.
Results: For training assessment, the images were divided into two regions – bone (HU >300) and soft tissue (HU<300). The MAEs from bone and soft tissue is 445±86 HU and 138±34 HU respectively with average MAEs of 192±32 HU. Histograms from both the pCT and sCT were similar. The shot times for TMR10 was consistently lower than CA for all sites ranging from 2.3% to 5.98% while the CA from sCT was closer to pCT with error ranging from -0.29 to 1.11%.
Conclusion: Dose calculated with CA from sCT is in better agreement with CA from pCT than TMR10. Thus, tissue inhomogeneity correction is possible with sCT based on MRI images acquired as standard of care. This negates the need for pCT acquisition which is dosimetrically and economically advantageous.