Reconstruction of 3D CT from A Single X-ray Projection View Using CVAE-GAN

Computed tomography can provide a 3D view of the patient's internal anatomy. However, traditional CT reconstruction methods require hundreds of X-ray projections through a full rotational scan of the body, which cannot be performed on a typical X-ray machine. In order to deal with the impact of organ movement caused by respiration in radiotherapy on the accuracy of radiotherapy, we propose to reconstruct CT from a single X-ray projection view using the conditional variational autoencoder. Conditional variational autoencoder encodes the features of a 2D X-ray projection. The decoder decodes the hidden variables encoded by the encoder and increase data dimension from 2D (X-rays) to 3D (CT) to generates a corresponding 3D CT. In addition, we use the discriminator to distinguish the generated 3D CT from the real 3D CT to make the generated 3D CT more realistic. We demonstrate the feasibility of the approach with 3D CT of two patients with lung cancer.

I. Introduction

The precise knowledge of tumor position is important for intra-operative image-guidance of various treatment. For example, in radiotherapy of lung cancer, the respiratory motion causes the change of the position of the tumor and surrounding tissues, which leads to the uncertainty of the radiation dose. It is very necessary to obtain the actual 3D anatomical information during the treatment to guide the treatment and analyse the actual dose distribution.