Uncertainty-Aware Multi-resolution Whole-Body MR to CT Synthesis

Synthesising computed tomography (CT) images from magnetic resonance images (MRI) plays an important role in the field of medical image analysis, both for quantification and diagnostic purposes. Especially for brain applications, convolutional neural networks (CNNs) have proven to be a valuable tool in this image translation task, achieving state-of-the-art results. Full body image synthesis, however, remains largely uncharted territory, bearing many challenges including a limited field of view and large image size, complex spatial context and anatomical differences between time-elapsing image acquisitions. We propose a novel multi-resolution cascade 3D network for end-to-end full-body MR to CT synthesis. We show that our method outperforms popular CNNs like U-Net in 2D and 3D. We further propose to include uncertainty in our network as a measure of safety and to account for intrinsic noise and misalignment in the data.