Scalable and modular online data processing for ultrafast computed tomography using CUDA pipelines

Abstract

For investigations of rapidly moving structures in opaque technical devices ultrafast electron beam X-ray computed tomography (CT) scanners are available at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR). Currently, measurement data must be initially downloaded after each CT scan from the scanner to a data processing machine. Afterwards, cross-sectional images are reconstructed. This limits the application fields of the scanners. For online observations and even automated process control of scanned objects a new modular data processing tool is presented consisting of user-definable pipeline stages that work independently together in a so called data processing pipeline that can keep up with the CT scanner's frame rate of up to 8 kHz. The data processing stages are arbitrarily programmable and combinable and are connected by a fast custom memory pool to optimize data transfer processes. As a result, this processing structure is not limited to CT application only. In order to achieve highest processing performances for the electron beam X-ray CT scanners all relevant data processing steps are individually implemented in separate stages using graphic processing units (GPUs) and NVIDIA's CUDA programming language. Data processing performance tests on two different high-end GPUs (Tesla K20c, GeForce GTX 1080) offer a slice image reconstruction performance that is well-suited for online application.