Real-time quasi-3D tomographic reconstruction

Abstract

Developments in acquisition technology and a growing need for time-resolved experiments pose great computational challenges in tomography. In addition, access to reconstructions in real time is a highly demanded feature but has so far been out of reach. We show that by exploiting the mathematical properties of filtered backprojection-type methods, having access to real-time reconstructions of arbitrarily oriented slices becomes feasible. Furthermore, we present RECAST3D, software for visualization and on-demand reconstruction of slices. A user of RECAST3D can interactively shift and rotate slices in a GUI, while the software updates the slice in real time. For certain use cases, the possibility to study arbitrarily oriented slices in real time directly from the measured data provides sufficient visual and quantitative insight. Two such applications are discussed in this article.

Real-time reconstruction and visualisation towards dynamic feedback control during time-resolved tomography experiments at TOMCAT

“…To achieve real-time visualisation of tomographic experiments, our pipeline includes two main parts: a detector component that provides direct access to acquired projections in real time, and a software component that can process the acquired data and visualise results in real time. In the realisation we present here, the detector component is implemented using the GigaFRoST detection and readout system 4 , while the software component consists of the RECAST3D real-time reconstruction and visualisation software and streaming architecture 12 . We will first discuss in more detail the elements of both components relevant to the presented pipeline, and then explain how the two components were integrated.…”

“…The RECAST3D framework 12 provides a quasi 3D reconstruction of the scanned object by simultaneously reconstructing and visualising a set of arbitrarily oriented tomographic slices, which can be dynamically chosen by the user and are constantly updated in real time (Fig. 1).…”

“…The quasi-3D reconstruction pipeline of RECAST3D 12 is built upon a message-passing protocol between a visualisation tool for reconstructed slices and a reconstruction server. The reconstruction server holds (preprocessed) tomographic projections in memory, and is able to reconstruct arbitrarily oriented slices from this data on demand, e.g.…”

“…Instead of computing an entire 3D snapshot of the scanned object, our approach computes multiple arbitrarily oriented slices. The pipeline is based on combining the GigaFRoST detector system 4 , which provides direct access to newly acquired projections, with the recently published RECAST3D software 12 , which enables real-time visualisation of arbitrary oriented slices by directly reconstructing the slices from the measured projections. The image reconstruction part of our pipeline runs on a single GPU-equipped workstation, thereby providing an imaging solution that can be implemented at the beamline in a straightforward manner without need for on-demand access to compute and network resources at a supercomputing facility.…”

Advanced x-ray tomography: experiment, modeling, and algorithms

“…Besides image quality, another crucial performance metric of the tomography pipeline is the time required for a scan and the computational requirements of its data processing. The articles [14][15][16] focus on various aspects of high-throughput tomography systems, related to running time and data storage requirements.…”