Investigations of an EPID-based 3D dose reconstruction method for applications in MRI-Linac radiotherapy

One of the recent developments in radiotherapy is the integration of MRI scanners with a radiation source, usually a linear accelerator (linac). These MRI-Linac systems allow for more accurate distinction between healthy tissue and tumour tissue. Electronic portal imaging devices (EPIDs), originally utilized for patient position verification, are now being increasingly used for dose verification in radiotherapy due to their fast image acquisition and high resolution, as well as the potential for in vivo measurements and 3D dose verification. In this thesis, a dose verification method utilizing a forward EPID dose prediction model and a 3D patient dose reconstruction algorithm is investigated for its potential use in the Australian MRI-Linac system. The dose prediction model and dose reconstruction algorithm were both previously developed by a group at the CancerCare Center in Manitoba, Winnipeg, Canada. The accuracy of the dose verification method was first tested on a conventional linac system, where a forward prediction model was calibrated for a 6X TrueBeam linac using a Varian as1200 EPID. Two square field plans and two clinical IMRT plans were delivered, with EPID images acquired during the deliveries. A MatriXX 2D ion chamber array detector was used to independently measure the dose delivered to the phantom. The forward prediction model was shown to predict the 2D dose delivered to the EPID with an accuracy > 95.8% for square field plans and > 99.8% for clinical IMRT plans, using a 2D gamma comparison with 3% dose difference and 2 mm distance to agreement criteria. The accuracy of the patient dose reconstruction algorithm was also evaluated with a 2D gamma comparison, comparing the reconstructed doses with measured MatriXX detector doses. The gamma comparisons yielded pass rates > 96.0% for square field plans, and > 93.2% for clinical IMRT plans, using a 3%/2 mm criterion. However, both the reconstructed doses and measured MatriXX doses showed much worse agreement with the treatment planning system doses, with gamma pass rates as low as 45.8% with the 3%/2 mm criterion. The forward EPID dose prediction model was investigated for transit dosimetry with MRI-only treatment planning, as this will be used in the Australian MRI-Linac system. The forward model was used to evaluate the accuracy of 8 clinical treatment plans that were performed as part of an MRI-only planning study. The patients were all CT scanned for quality assurance, and also had sCT scans generated using MRI scans. The forward EPID prediction model was used to predict the dose delivered to the EPID, using both the CT dataset and the sCT dataset, and compared to the measured EPID images acquired at the first fraction of the treatments. The average gamma pass rates for the 5%/2mm criterion were 90.3% ± 2.0% and 89.9% ± 2.0% for the sCT and CT predictions respectively. For the 5%/1mm criterion the average gamma pass rates were 85.0% ± 2.4% and 83.5% ± 2.2%, for the 4%/2mm criterion the average gamma pass rates were 85.9% ± 2.5% and 84.8% ± 2.3%, and for the 3%/2mm criterion the average gamma pass rates were 80.2% ± 2.9% and 77.8% ± 2.7% for the sCT and CT predictions respectively. Comparing the TrueBeam linac predictions on their own, the gamma pass rates increased to 99.9% ± 0.0% and 99.6% ± 0.0% for the 5%/2mm criterion, 99.3% ± 0.0% and 98.8% ± 0.0% for the 5%/1mm criterion, 99.7% ± 0.0% and 98.9% ± 0.0% for the 4%/2mm criterion, and 98.7% ± 0.0% and 97.3% ± 0.0% for 3%/2mm for the sCT and CT predictions respectively. To evaluate the suitability of EPID dosimetry for use in the Australian MRI-Linac system, the dosimetric properties of an amorphous silicon (a-Si) EPID operating in the Australian MRI-Linac system were investigated. A PerkinElmer a-Si EPID was used to acquire measurements to investigate the EPIDs field size response, beam symmetry and flatness, linearity with dose, signal response reproducibility, ghosting/image lag effects and the EPIDs response during MRI image acquisition. The field size response, linearity with dose and beam symmetry and flatness were all comparable to an EPID operating in a conventional linac system. The EPIDs response reproducibility did show variation, but this was found to be caused by variations in the linac output. Once this linac output variation was accounted for, the EPIDs response reproducibility was improved. The ratio of dark field images taken before and during MRI image acquisition yielded a mean value of 1.0001, with standard deviation equal to 0.1% of the mean. The ratio of two 3 x 3 cm² fields, delivered with a dose of 5 GSF, taken before and during MRI image acquisition yielded a mean value of 0.9667, with a standard deviation equal to 0.16% of the mean. These differences occurred almost entirely at the base of the penumbra of the delivered field, which is typically a low dose region. For the area inside the field the differences were all within ± 5%. Overall, it was concluded that an a-Si EPID was suitable to be used for dose verification in the Australian MRI-Linac system. A forward EPID dose prediction model was created for the Australian MRI-Linac system, calibrated using EPID images measured on the system. The model was adapted to account for certain differences presented in the MRI-Linac system compared with conventional linac systems. The 2D gamma comparisons between the forward prediction model and measured EPID images of open square fields resulted in pass rates > 70.2% using a 3%/2mm criterion, with an average pass rate of 87.0%. This agreement was found to decrease when the thickness of phantom in the beam increased, with the 2D gamma pass rates falling as low as 45.7% for a 1 x 1 cm² field attenuated through a 20 cm thick solid water phantom. The comparison between the reconstructed dose and the Farmer chamber measurements showed no difference in CAX dose, but did display field size differences of 5.5% in the crossplane direction and 0.6% in the inplane direction. For the film measurements taken horizontally along the direction of the beam, the reconstructed doses had an average 2D gamma pass rate of 73.3% across all field sizes for the 3%/2mm criterion, with a minimum pass rate of 54.6%. The measured film images were cropped to exclude the first 2 cm of phantom depth to account for the skin dose increase caused by the inline magnetic field that was not included in the prediction model or dose reconstruction algorithm. After the images were cropped, the reconstructed doses had an average 2D gamma pass rate of 78.9% across all field sizes measured for the 3%/2mm criterion, with a minimum pass rate of 53.2%. The 3D patient dose reconstruction algorithm was shown to be capable of reconstructing the dose delivered to the patient in the Australian MRI-Linac system with an accuracy > 54.6% for 2D gamma comparisons with a 3%/2mm criterion. Further improvements to the prediction model and dose reconstruction algorithm must be made before it can be clinically implemented with the Australian MRI-Linac system. The modelling of the inplane penumbras of the fields will need to be improved, and the concentrated electron contamination from the inline magnetic field will need to be modelled.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.