Fast online Monte Carlo-based IMRT planning for the MRI linear accelerator.

The MRI accelerator, a combination of a 6 MV linear accelerator with a 1.5 T MRI, facilitates continuous patient anatomy updates regarding translations, rotations and deformations of targets and organs at risk. Accounting for these demands high speed, online intensity-modulated radiotherapy (IMRT) re-optimization. In this paper, a fast IMRT optimization system is described which combines a GPU-based Monte Carlo dose calculation engine for online beamlet generation and a fast inverse dose optimization algorithm. Tightly conformal IMRT plans are generated for four phantom cases and two clinical cases (cervix and kidney) in the presence of the magnetic fields of 0 and 1.5 T. We show that for the presented cases the beamlet generation and optimization routines are fast enough for online IMRT planning. Furthermore, there is no influence of the magnetic field on plan quality and complexity, and equal optimization constraints at 0 and 1.5 T lead to almost identical dose distributions.

WE-G-BRCD-08: Virtual Couch Shift (VCS) by Online Plan Re-Optimization for the MRI Linear Accelerator.

PURPOSE: With the MRI accelerator it will be possible to get continuous patient anatomy updates, ranging from organ deformation to patient translation. To compensate for translations, one can re-optimize the treatment plan based on the online MRI images. Consequently, the IMRT optimization system should be fast and robust enough to generate daily a clinically acceptable plan to perform this 'virtual couch shift (VCS)'. METHODS: The system uses a GPU based Monte-Carlo dose engine (GPUMCD) for online beamlet generation in a 1.5 T magnetic field and a fast inverse dose optimization algorithm (FIDO). For four phantom and two clinical cases (cervix and kidney), we generated clinically acceptable plans. The given plans are regenerated after a series of x, y, and z translations (up to 34 mm) of the patient anatomy, without adapting the optimization constraints as used during the initial optimization. The differences between the original plan and the regenerated plans are evaluated by using the gamma criterion and the relative D99 target coverage. RESULTS: The system accurately reproduced the initial dose distribution after translating the phantom and patient anatomies. The gamma criterion of 2%/2 mm is satisfied for 99.2% all target voxels and for 97.2% for all OAR voxels. The relative D99 differences are almost 0.0 with a small standard deviation. With current hardware, a 7 beam cervix beamlet generation and IMRT optimization takes 141 seconds, the kidney case takes only 14 seconds. CONCLUSIONS: We developed a system which is fast and accurate enough to perform a VCS by online re-optimization for the MRI accelerator. Currently we are adding sequencing to the system. We expect that this method can also be used for compensating patient rotations and tissue deformation and with this go towards realtime adaptive treatment planning and delivery.

Fast dose calculation in magnetic fields with GPUMCD.

A new hybrid imaging-treatment modality, the MRI-Linac, involves the irradiation of the patient in the presence of a strong magnetic field. This field acts on the charged particles, responsible for depositing dose, through the Lorentz force. These conditions require a dose calculation engine capable of taking into consideration the effect of the magnetic field on the dose distribution during the planning stage. Also in the case of a change in anatomy at the time of treatment, a fast online replanning tool is desirable. It is improbable that analytical solutions such as pencil beam calculations can be efficiently adapted for dose calculations within a magnetic field. Monte Carlo simulations have therefore been used for the computations but the calculation speed is generally too slow to allow online replanning. In this work, GPUMCD, a fast graphics processing unit (GPU)-based Monte Carlo dose calculation platform, was benchmarked with a new feature that allows dose calculations within a magnetic field. As a proof of concept, this new feature is validated against experimental measurements. GPUMCD was found to accurately reproduce experimental dose distributions according to a 2%-2 mm gamma analysis in two cases with large magnetic field-induced dose effects: a depth-dose phantom with an air cavity and a lateral-dose phantom surrounded by air. Furthermore, execution times of less than 15 s were achieved for one beam in a prostate case phantom for a 2% statistical uncertainty while less than 20 s were required for a seven-beam plan. These results indicate that GPUMCD is an interesting candidate, being fast and accurate, for dose calculations for the hybrid MRI-Linac modality.