sCT and Dose Calculation
Résumé
Magnetic resonance imaging (MRI) has recently established itself as a new standard in radiatiotherapy, owing to its high soft tissue contrast enabling a significantly more accurate volume segmentation and better characterisation of anatomical changes during treatments. The underlying synthetic computed tomography (sCT), required in clinics implementation for dose computation, is extensively investigated in this chapter. First, the generation methods, including bulk density assignment, atlas-based and voxel-based approaches, as well as the associated pros/cons, are described. An appealing compromise between ease, efficiency, and speed is bulk density assignment, which is already implemented in one commercial MRI-Linac. Very recently, however, deep learning has gained the upper hand and is set to become the reference method in clinical practice in the very near future. Second, the metrics to perform a multi-criteria sCT image quality evaluation are provided, as well as the latest performance obtained in the literature. High interest metrics include the body mean absolute error, dose volume histograms differences, global gamma indices with low-/high-dose thresholds, and metrics characterising registration differences between online positioning images and sCT/CT images. These metrics are complementary and enable to respectively assess Hounsfield units recovery, organ-scale dosimetric agreement, global dosimetric agreement in low-/high-dose regions, and patient setup ability. Third, after a reminder of some of the basics of distortions and artefacts in MR imaging, the latest recommendations in terms of assurance quality are described, with the ultimate aim of maximising the quality of the sCT produced. A particular focus is made on B0 inhomogeneities, residual gradient non-linearity, and susceptibility artefacts, owing to their high occurrence in clinical routine. Lastly, a concrete literature review of commercially available sCT products implementations into clinics, either with conventional linear accelerators (Linac) or with hybrid MRI-Linac, is provided. The associated performance, based on the metrics described in the second section, are also included.