Patient selection
Patients with localized prostate cancer were included in this study from March 2017 to May 2018. All patients were referred to RT of the prostate gland alone with patient set-up verification based on GFM, with MR- and CT-imaging as a part of their prescription. Patients with metallic implants in the pelvic area or any MR-contraindications were not considered for inclusion.
MRI-only workflow
The MRI-only workflow used in this study was inherited from the dual-modality workflow for prostate cancer patients in our clinic. The goal was to keep the same structure as the previous dual-modality workflow and make adjustments when required for MRI-only. The MR-PROTECT MRI-only workflow is schematically presented in Fig. 1 and its content explained in the figure text.
To enable prospective and retrospective analysis of the workflow, CT-images were acquired for each patient. The CT-images were imported to the clinical systems after planning approval of the MRI-only treatment plan and were not available in the treatment planning system (TPS) during any task prior to the CT-QA. Electronic questionnaires were introduced in the MRI-only workflow using ARIA (v.13.6, Varian Medical systems, Palo Alto, California, USA) to guide the clinical staff during the implementation. If problems were detected during the workflow, a patient could if necessary be converted to the conventional dual-modality workflow. All questionnaire items are found below the respective task (green check marks in Fig. 1). The tasks in the workflow are presented in the following subsections.
Imaging
MR-imaging
The MR-scanner used was a GE Discovery 750 W 3.0 T (Software version DV25.1-R02–1649.a, GE Healthcare, Chicago, Illinois, USA). The MR-QA was performed according to clinical practice during the study, which included a monthly geometric distortion check using a large field of view (FoV) phantom (GRADE, Spectronic Medical AB, Helsingborg, Sweden) [23, 24].
The MR-protocol consisted of a large FoV (L-FoV) T2 weighted (T2w) sequence, primarily used for sCT-generation, target and OAR delineation and GFM definition. Three small FoV (s-FoV) T2w sequences (transverse, coronal and sagittal projections) were acquired for target delineation support and a multi-echo gradient echo (MEGRE) sequence for GFM-identification [25]. The L-FoV images were acquired between the MEGRE and the transversal s-FoV sequence in the sequence order to minimize impact of intra MR-protocol patient motion. To minimize the impact of geometric distortion large receive bandwidth per pixel, 2D together with 3D gradient distortion correction and automatic volume shimming were used. The MR-protocol, sequence parameters and order are presented in Appendix A. Since all tasks were performed primarily in the L-FOV image geometry, no image registration between the different MR-images were needed.
Patients were immobilized with ankle and knee support and scanned on a flat table top using a 16 channel GE GEM Anterior Array coil, positioned on stiff coil bridges. Patient tattoos for user-origin definition and patient RT alignment were created at the MR-scanner. To enable visualization of the tattoos and their locations in the L-FOV MR-images, cone shaped liquid surface markers (Pin Point for image registration 128, Beekley Medical, Bristol, Connecticut, USA) were used.
CT-imaging
The CT-imaging was performed directly after the MRI using a Siemens Somatom Definition AS+ (Software version syngo CT VA48A, Siemens Healthineers, Erlangen, Germany) with 3 mm slice thickness and a tube voltage of 120 kV. Patients were positioned as during the MRI, aligned using the tattoos defined at the MR-imaging task. The CT-images were strictly used for QA purposes and retrospective analysis and were imported into the TPS during the CT-QA task.
Post imaging
MR-parameter check
To minimize the risk of an unintentional change in MR-parameters, experienced previously [22], a MATLAB script (v. 2015b, Mathworks Inc., Natick, MA, USA) for MR-parameter check was developed and used. The script compared the MR-parameters in the L-FoV image against a predefined template and notified the user by e-mail regarding compliance to the template or not (source code available at: https://github.com/jamtheim/MRIAcqParameterCheck). Any deviations and corresponding reasons were recorded and further evaluated.
sCT-generation
The sCT-generation (MriPlanner™ v.1.1.2, Spectronic Medical AB, Helsingborg, Sweden) has earlier been described and multi-center/multi-vendor validated [22, 26]. The cloud based solution is connected through a DICOM export node on the MR-scanner, facilitating automatic sCT-generation when the L-FoV images are sent to the node. The sCT-images were automatically returned to the TPS (Eclipse v. 13.6, Varian Medical systems, Palo Alto, CA, USA), DICOM-modality-tagged as a CT and placed in the same frame of reference (FoR) as the MR-images.
GFM identification
Three cylindrically shaped GFMs (1 mm diameter × 5 mm long) were implanted into the prostate using a biopsy needle 2 weeks prior imaging. The GFMs resulted in signal voids in both the L-FOV and the MEGRE-images. GFMs were identified using the MEGRE-images, which has been described earlier [25], and their spatial positions were manually identified in the L-FoV images. A DICOM-viewer (MicroDicom v.2.7.9, MicroDicom, Sofia, Bulgaria) was used to display the MEGRE-images during GFM-identification. Two operators identified the GFMs in 24 and 11 patients respectively and 5 together. The spatial positions of the GFMs defined as a RT-structure in the L-FoV images were exported to the MriPlanner™ from the TPS, enabling creation of synthetic GFMs in the sCT-images. This required that the GFMs spatial position was defined in a physical slice (and not in between) in the L-FOV images. The synthetic GFMs were represented in one image slice each in the sCT as round high intensity objects with a diameter of 4 mm.
Treatment plan creation
Target and OAR delineation
Delineation of target and OARs was performed based on the L-FoV images, supported by the s-FoV images. A blended view with the different MR-images was used to define the structures in the L-FoV image geometry. The sCT-images were not used for delineation, but the final structures set were created in the sCT geometry (i.e. the L-FoV geometry).
There was no intention to change the MRI-based planning target volume (PTV) compared to the dual-modality workflow. Therefore, based on an earlier study, a 1 mm extra margin (excluding cranio-caudal extension) was added to the clinical target volume (CTV) to compensate for the smaller MRI-based CTV [27]. A 7 mm isotropic margin was used to create the MRI-based PTV according to clinical practice.
Treatment planning
A 10 MV volumetric modulated arc therapy (VMAT) treatment plan was created using the sCT for each patient according to clinical practice. Dose prescription was 78 Gy in 39 fractions. The standard HU to ED curve defined in our clinic was used for dose-calculations. The dose-calculations were performed using an Analytical Anisotropic Algorithm (AAA) (v.13.6.23, Varian Medical systems, Palo Alto, California, USA). Minor modifications to the standard treatment planning procedure were needed with respect to the use of a sCT in our TPS. This included a manually inserted and positioned treatment couch structure at a 1.5 cm distance below the sCT body contour, reflecting the thickness of the mattress used during treatment. The treatment user-origin was defined using the liquid markers placed over the tattoos at the MR-imaging task. Digitally reconstructed radiographs (DRR) for patient positioning were created from the sCT at a gantry angel of 0 and 270 degrees.
Quality assurance
After import to the TPS, the CT-images were automatically rigidly registered based on the bony anatomy in translational directions to the sCT-images. The treatment plan transfer was performed in translational directions only, disregarding any rotations in the registration. Hence, the registration between sCT and CT was also performed in translational directions only. Separate body contours were created for both images. The sCT treatment plan was recalculated on the CT-images using the same field setup and number of monitor units. Deviations between sCT and CT dose-distributions were evaluated based on PTVmean directly in the TPS by transferring the PTV structure to the CT. From experiences in earlier validation studies [22], a PTVmean dose deviation less or equal to 1% for each patient was considered acceptable. Deviations above 1% were further investigated for approval depending on the reasons and magnitude of the deviations. A qualitative comparison of the general appearance of the HU in the sCT- and CT-images was performed using HU-line profiles in the TPS. Verification of the GFMs was performed with visual sanity assessment of the GFMs positions in the sCT- and CT-images and corresponding DRRs. All questionnaire items (Fig. 1) had to be acknowledged to enable MRI-only treatment approval.
After approval of the MRI-only treatment plan, our standard clinical QA was performed including a verification measurement using the Delta4 phantom (Scandidos, Uppsala, Sweden) and gamma analysis [28]. The clinical gamma analysis was performed using a 3%/2 mm global criteria and a 15% dose cut-off, with a minimum pass rate of 90% comparing the measured and planned dose.
Treatment delivery
When all QA-steps were completed, the MRI-only treatment plan was approved and passed on to daily treatment on a TrueBeam accelerator (Varian Medical systems, Palo Alto, California, USA). Patients were positioned with corresponding fixation as during MRI and aligned using the patient tattoos. Set-up verification was performed with daily kilo voltage (kV)-image registration. The synthetic GFMs, represented in the sCT-DRR, were manually registered towards the physical GFMs seen in the orthogonal kV-image pairs. Operators were instructed to match the center of the GFM in the orthogonal kV-images to the center of the corresponding synthetic GFM in the sCT-DRR. From the eleventh patient and forward, the cylindrical GFM shape was added to the sCT as a RT structure around the synthetic GFMs. This was an attempt to facilitate easier detection of prostate and GFM rotations.
Retrospective investigation
Analyses of the study population were performed to investigate achievable limits of acceptance criteria. This was performed retrospectively and included three tasks in the workflow; 1) GFM identification 2) treatment planning and 3) set-up verification. Each analysis and its method are presented below.
GFM identification
The GFM-identification performance was investigated according to the analysis presented by Greer et al. [15] where the common center of mass centroid of all sCT and CT-GFMs were determined and the distances from each GFM to their respective common centroid were calculated. The spatial location of each GFM center of mass (CoM) was determined in the sCT- and CT-images respectively using the method previously described by our research group in section 2.C of the paper by Gustafsson et al. [25]. The absolute distances from each GFM to the common centroid were compared between sCT- and CT-images.
This resulted in three GFM comparisons for each of the 40 included patients, given as a difference in mm, and hence a total of 120 comparisons. Mean GFM deviation for all comparisons was calculated along with the range, root mean square (RMS) and standard deviation (SD). The presence of intra-prostatic calcifications ≥2 mm in the CT-images were measured and noted.
Treatment planning
A comparison between the sCT and CT dose-distribution was performed for the DVH-criteria specified in the clinical protocol in our clinic. This included CTV Dmin, PTV D98 and D95, rectum D10, D15 and D30, Femoral head D2 and Bladder Dmean. Further, Dmean was extracted for all targets and OAR. The deviations between the two dose-distributions were calculated as difference in % of the prescribed dose. A gamma analysis was performed using global gamma criteria of 3%/3 mm, 2%/2 mm, 2%/1 mm and 3%/2 mm. A 15% dose cut-off was used and corresponded to the cut-off value used in our clinical gamma analysis. All evaluations were performed using MICE Toolkit (MICE Toolkit™, v.1.0.9, Nonpi Medical, Umeå, Sweden). The translational registration created between sCT- and CT-images in the prospective CT-QA was applied to the CT dose-distribution and was re-sampled to the sCT spatial resolution. The delineated MR-structures were used in the DVH-evaluation of both dose-distributions.
Set-up verification
An evaluation of patient set-up using sCT DRRs with synthetic GFMs was performed. The registrations were performed in the Offline Review module in the TPS by one operator. The sCT-DRR and CT-DRR were manually rigidly matched respectively towards orthogonal kV-image pairs acquired during the first three treatments for each patient. This resulted in the registrations sCT-kV and CT-kV. Registrations were performed according to the clinical method where the center of the GFM in the kV-images are manually aligned with the center of the corresponding synthetic GFM in the DRRs.
A CBCT-based bone match strategy was also evaluated. The sCT and CT-images were registered by one operator in the registration module in the TPS using auto-match towards CBCT-images acquired during one fraction for each patient. The auto-match anatomy was defined using a box, including the common bony anatomy in the images. A HU range of 200 to 1700 within the box was used for the match. This resulted in the registrations CT-CBCT and sCT-CBCT. As the clinical set-up verification did not include any rotations in the couch positioning of the patient, the retrospective registration was performed without rotations.
Differences in couch translations in x, y and z directions for the sCT-kV and the CT-kV registrations were compared for the first three treatment fractions for the first nine treated patients. The couch translation differences in x, y and z between the CT-CBCT and sCT-CBCT registrations were compared for all patients.
