🤖 AI Summary
Conventional CT reconstruction assumes static object geometry, rendering it ill-suited for in situ dynamic experiments and prone to motion artifacts and morphological distortion. To address 4D (3D spatial + temporal) dynamic CT reconstruction, this work proposes Distributed Implicit Neural Representation (DINR): it employs sparse continuous spatiotemporal coordinate sampling and distributed stochastic gradient optimization to circumvent memory explosion associated with voxel-grid discretization; integrates a differentiable CT forward projection model with multi-GPU cooperative training to enable large-scale dynamic deformation modeling. Evaluated on both simulated parallel-beam and real cone-beam CT data, DINR substantially suppresses motion artifacts, improves reconstruction accuracy by over 35%, and reduces GPU memory consumption by an order of magnitude. It represents the first approach to achieve high-fidelity, low-overhead, and scalable implicit 4D CT reconstruction.
📝 Abstract
4D time-space reconstruction of dynamic events or deforming objects using X-ray computed tomography (CT) is an important inverse problem in non-destructive evaluation. Conventional back-projection based reconstruction methods assume that the object remains static for the duration of several tens or hundreds of X-ray projection measurement images (reconstruction of consecutive limited-angle CT scans). However, this is an unrealistic assumption for many in-situ experiments that causes spurious artifacts and inaccurate morphological reconstructions of the object. To solve this problem, we propose to perform a 4D time-space reconstruction using a distributed implicit neural representation (DINR) network that is trained using a novel distributed stochastic training algorithm. Our DINR network learns to reconstruct the object at its output by iterative optimization of its network parameters such that the measured projection images best match the output of the CT forward measurement model. We use a forward measurement model that is a function of the DINR outputs at a sparsely sampled set of continuous valued 4D object coordinates. Unlike previous neural representation architectures that forward and back propagate through dense voxel grids that sample the object's entire time-space coordinates, we only propagate through the DINR at a small subset of object coordinates in each iteration resulting in an order-of-magnitude reduction in memory and compute for training. DINR leverages distributed computation across several compute nodes and GPUs to produce high-fidelity 4D time-space reconstructions. We use both simulated parallel-beam and experimental cone-beam X-ray CT datasets to demonstrate the superior performance of our approach.