Sparse patch-based method applied to mri-only radiotherapy planning

Abstract

Introduction Replacing CT/MRI in radiotherapy chain with MRI-only will remove CT/MRI fusion uncertainties and avoid the patient the dose of ionizing radiation from CT. Purpose We propose to generate pseudo-CT (pCT) from conventional MRI and to assess it for MRI-only brain radiotherapy planning (RTP). Material/methods In twelve patients with brain tumors, CT and contrast-enhanced T1-weighted MRI were registered. A library of patches was built by extracting 2D patches, defined as MRI squares of 5 × 5voxels, in each voxel location and labelling them with corresponding HU values. Each patch in target MRI was reconstructed from a sparse linear combination of database patches locally searched. The sparse coefficients were estimated by optimizing the Elastic-Net (EN) objective function. EN was enhanced by additional penalty term based on the structural similarity between target and database patches. Radiological and dosimetric assessments of pCT were done for all patients using leave one out cross-validation. Mean absolute error between CT and pCT radiological paths (MAE WEPL ) over a grid were computed. VMAT planning was performed on CT and pCT for PTVs delineated in homogenous (PTV1) and heterogeneous regions (PTV2) and compared. Percentage of dose metrics deviations (PDMD) for PTVs and OARS were computed. Results Radiological path estimation was accurate with MAE WEPL = 2.6 ± 0.4 mm. Good agreement with conventional planning techniques was obtained; the highest PDMD were D 98% = 0.05 ± 0.1 for PTV1, D 98% = −0.81 ± 0.47 for PTV2 and D mean = 0.35 ± 0.7 for left lens. Conclusion We presented a novel technique to generate pCT for MRI-only brain RTP. In the future, we will investigate its use for PET/MR attenuation correction.