Feasibility Study of MRI-Only Simulation for Cyberknife Skull Tracking
Abstract
Purpose
In an MRI-only simulation workflow, synthetic CT (sCT) are created from MR images and mimic the CT in terms of HU quality and geometric accuracy, therefore enabling dose calculation and patient setup verification. The elimination of CT acquisition leads to less radiation to the patient and improved workflow efficiency. The purpose of this work is to determine whether sCT-based DRR images are suitable for treatment of intracranial lesions with CyberKnife Skull tracking.
Methods
T1 Dixon MRI (in-phase, out-of-phase, water, and fat images) of the brain were acquired on a 3-Tesla scanner (MAGNETOM Vida, Siemens Healthineers). Synthetic CT images were created using a commercially available 2D deep learning-based prediction model (Syngo.via, Siemens Healthineers). DRRs were generated from the sCT using the Accuray Precision TPS and the same geometry/align center as the corresponding CT-based DRRs. 210 images from 3 patients were analyzed using 10 image similarity metrics; including mean squared and absolute errors, person correlation coefficient (PCC) and normalized mutual information (NMI). In addition to these metrics, the image sharpness was also calculated using the gradient strength map.
Results
CT and sCT-based DRRs present a moderate correlation with NMI of around 1.20+-0.01 and PCC of 53.5+-2.0%. The image sharpness result was of 0.118+-0.010 for the CT whereas sCT had 0.148+-0.011.
Conclusion
Global similarity metrics reveal small differences between the CT-based DRR and sCT-based DRR, particularly at the sinus cavity (where strong bone/tissue/air susceptibility differences can pose challenges for MR-based sCT reconstruction) and the posterior neck (where there can be small differences in neck flexion). Image sharpness metrics, which highlight bone and tissue differentiation, seem to strongly support the usability of sCT for skull tracking, with sCT significantly outperforming CT images. Future work is needed to test the delivery of the plan on the machine using anthropomorphic phantoms.