Room: Track 2
Purpose: In current spine SBRT practice, patient positioning is often re-verified after every treatment arc, guided by repeated on-treatment CBCT scans. A method is presented to automatically detect patient position changes that occurred during an arc, thereby reducing the number of required CBCT scans and accelerating treatment workflow.
Methods: The proposed method uses an initial, dual-energy set-up CBCT scan to generate a virtual mono-energetic image (VMI) at the higher energy of the treatment beam. The VMI is then aligned via rigid 3D-2D registration with the electronic portal imaging device (EPID) data captured in each arc. This determines a best fit translation of the bed needed to realign the patient with the treatment beam. Moreover, because the VMI and the treatment beam are at the same energy, a simplified sum-of-squared-differences (SSD) similarity criterion can be used for the 3D-2D registration step. The method was tested on a simulated treatment of a spine lesion in a digital thorax phantom. Two different multi-leaf collimator (MLC) leaf sequences were assessed, one which enveloped the entire affected vertebra body and one which excluded the spinal cord. For each case, the simulated EPID data included a 0.9 cm patient shift relative to the set-up CBCT. The proposed registration method was applied to estimate the patient misalignment. Performance was quantified in terms of the 3D displacement between the estimated and ground truth patient position.
Results: As expected, the tighter leaf sequence resulted in larger registration error. For both sequences however, sub-millimeter positioning accuracy was achieved using approximately 30 EPID frames.
Conclusion: These preliminary tests show that dual-energy VMIs may be registered to MLC-collimated EPID data to give accurate estimates of patient position changes. This in turn promises to accelerate treatment and to reduce on-treatment CBCT dose in SBRT procedures.
Image-guided Therapy, Registration