Room: Stars at Night Ballroom 2-3
Purpose: The advent of MRI-based therapy delivery systems is driving an interest in MRI-guided radiation therapy (MR-IGRT). MRI achieves superior soft tissue contrast compared to x-ray CT, allowing for applications like online adaptive therapy in pursuit of improved outcomes. The expanding use of MR-IGRT makes an MR-only RT workflow in which MRI is the sole imaging modality used for treatment planning and dose calculations an attractive goal. Such a workflow is enabled by the generation of synthetic CT (sCT) data that is used in dose calculations. We present the application of our deep spatial pyramid convolutional framework to the sCT reconstruction task in a preliminary clinical workflow for MR-only breast RT.
Methods: We view sCT reconstruction as an image-to-image translation task in which sCT images are generated directly from a patientâ€™s MR images. The framework developed and implemented here utilizes atrous convolution in a generative model characterized by a U-Net architecture. In this preliminary MR-only clinical workflow, the patient skips the conventional CT simulation, undergoing only a 1.5 T mDixon MRI simulation in the treatment position with the associated devices. Following simulation, sCT data is generated and exported using the trained model packaged in an in-house GUI. The proposed workflow was evaluated alongside the conventional workflow in which CT simulation is performed.
Results: Calculated dose metrics for the sCT-based plans demonstrate good agreement with those of the conventional plans for 6 test patients that have been evaluated at this point. Passing rates of greater than or equal to 98% are observed using the 3D gamma index with 2%/2 mm criterion.
Conclusion: The preliminary clinical workflow for MR-only RT of the breast presented here is enabled by our deep spatial pyramid convolutional framework. Such a workflow is an important step in taking full advantage of the applications of MR-IGRT.