Implementation of an MR-Only radiotherapy workflow for CNS: from commissioning to clinical use
MO-0223
Abstract
Implementation of an MR-Only radiotherapy workflow for CNS: from commissioning to clinical use
Authors: Sevgi Emin1, Elia Rossi2, Giovanna Gagliardi1,3, Mattias Hedman2,3, Fernanda Villegas1,3
1Karolinska University Hospital, Medical Radiation Physics and Nuclear Medicine, Stockholm, Sweden; 2Karolinska University Hospital, Department of Radiation Oncology, Stockholm, Sweden; 3Karolinska Institute, Department of Oncology-Pathology, Stockholm, Sweden
Show Affiliations
Hide Affiliations
Purpose or Objective
Magnetic Resonance (MR) images are widely used alongside computed tomography (CT) images for radiotherapy (RT) planning, introducing geometrical uncertainties due to the image registration process. An MR-only workflow avoids these uncertainties by eliminating the CT step and providing a synthetic CT (sCT) image which is generated directly from MRI.
This study describes the clinical implementation process, from commissioning (retrospective study) to validation (prospective study) of the commercial sCT product MRCAT brain (Koninklijke PHILIPS N.V.). The artificial intelligence software is now used clinically in an MR-only workflow designed for glioblastoma patients in our center.
Material and Methods
A total of 100 CNS patients were enrolled in the commissioning phase, where a specific sequence was acquired for the generation of the sCT, alongside routine imaging for CT based RT workflow. The contouring capability of the sCT was evaluated by comparing clinical target volumes (CTV) and normal tissue delineated on sCT/MR images versus routine delineation on CT/MR. Dose planning performance was evaluated for 24 glioma patients by dosimetric comparison of relevant DVH points.
A set of quality control (QC) guidelines was developed based on the results of the commissioning phase to assess sCT image and dose plan quality. Once in place, 24 glioblastoma grade IV patients were enrolled in the validation phase, receiving treatment with the MR-only workflow. A backup CT simulation was taken in case sCT failed the QC. Retrospective dose comparison between CT and sCT was performed to verify dosimetric agreement seen during the commissioning phase.
Results
Comparison of CT and sCT based structures yielded a mean Dice coefficient of 0.95 ± 0.02 and 0.93 ± 0.05 for brain and CTV structures respectively. The average Hausdorff distance was calculated to be less than 0.05 mm for both structures.
Dosimetric comparison between CT- and sCT-based RT plans showed an average dose difference of 0.60% in the mean dose to CTV, both for the commissioning and validation data. The average difference in the dose to normal tissue was less than 1.90% for the measures shown in figure 1. However, the interpatient variation was larger for the normal tissue compared to the CTV with dose differences up to 8.90% for an individual patient.
Bone artefacts were observed in post-operative patients with metal implants, which led to an erroneous thickening of the artefact in the sCT. Dose comparison for cases where the CTV was located adjacent to the metal implant, showed no impact on the mean dose.
Conclusion
The dosimetric results obtained in the commissioning and validation process of the commercial sCT product demonstrated that it can be used for RT planning with equal clinical confidence as in CT/MR based workflow. Moreover, the large cohort of patients in the study provided a big enough sample of image artefacts, which in turn facilitated the development of a robust QC program for MR-only workflow.