Dual-energy CT for automatic organ at risk segmentation in brain tumor patients using a multi-atlas and deep-learning approach

In radiotherapy, computed tomography (CT) datasets are frequently used to calculate radiation treatment plans to interpret dose evaluation metrics in healthy surrounding organs that need to be spared, the organs at risk (OARs). Based on CT scan and/or magnetic resonance images, the OARs have to be delineated by hand which is one of the most time-consuming tasks in the clinical radiotherapy workflow. Recent multi-atlas (MA) or deep learning (DL) based methods aim to improve the clinical workflow by automatically segment the OARs on a CT dataset. However, no studies investigated the performance of these MA or DL methods on dual-energy CT datasets which have been shown to improve the image quality compared to 120 kVp single-energy CT. In this study, the in-house developed MA method and the DL method (two-step three dimensional U-net) are described first. Then, the performance of both approaches (MA and DL) was quantitatively and qualitatively evaluated on various dual-energy CT datasets, more specifically on pseudo-monoenergetic CT dataset that were generated between 40 keV and 170 keV.