Spatially-Consistent Implicit Volumetric Function for Uni- And Bi-Planar X-Ray-Based Computed tomography Reconstruction

We introduce a spatially-consistent implicit function representation (sci-f) for high-resolution volumetric computed tomography (CT) image recovery from uni- and bi-planar X-ray images. We devise a deep end-to-end learning scheme to parameterize the unified implicit function conditioned on the input 2D X-rays and predict the detailed 3D anatomies. Instead of the discretized voxel representation in the existing deep learning-based CT reconstruction from sparse 2D X-rays, the lightweight and memory-efficient sci-f enables a volume recovery with continuous resolutions. The sci-f addresses the consistency in the projective plane and bridges the 2D pixels with their counterpart voxel through the 3D ray for spatially consistent 3D reconstruction of anatomical structures. Extensive experiments demonstrate the efficacy of the proposed approach in X-ray-based volume recovery from clinically obtained X-rays. The proposed approach has improved performance over state-of-the-art deep X-ray-based CT reconstruction.