A multi-channel reconstruction toolkit for computed tomography

Abstract According to the FDA, the recent clinical availability of photon counting detector X-ray CT (PCCT) marks the “first major imaging device advancement for computed tomography in nearly a decade” because it provides fundamental improvements in terms of image noise, spatial resolution, material discrimination, and low contrast

detectability. Yet this new era of clinical PCCT follows years of preclinical research at our Quantitative Imaging and Analysis Lab at Duke University and our preliminary research projects using the NAEOTOM Alpha clinical PCCT scanner from Siemens and installed at Duke. To support these research projects and under the Aims of

our funded NIA grant (1RF1AG070149; Cardiac photon counting CT and its application in studying interactions between Alzheimer's and heart disease), we are developing a GPU-based Multi-Channel Reconstruction (MCR) Toolkit for CT (channels refer to dynamic and/or multi-energy images of 3D anatomical structures). In addition

to being built from the ground up to handle the unique challenges of multi-channel reconstruction, our Toolkit includes several unique features not in other open-source packages: support for translational research using the same code on both clinical and preclinical data, robust reference algorithms for performing iterative

reconstruction of low-dose multi-channel CT data, and automatic data adaptation of reconstruction parameters to reduce manual parameter tuning. The objectives of this supplement are (1) to make the MCR Toolkit codebase more robust and extensible for CT experts and (2) to provide a graphical user interface and reference

reconstruction protocols for those using CT to conduct other research studies. We will accomplish these objectives through three distinct aims. In Aim 1 we propose to restructure our Toolkit codebase from its current C code to object oriented code written in C++, providing native support for popular file formats, eliminating

redundant code, reducing opportunities for user errors, and allowing reconstruction pipelines to be saved and shared. Aim 2 focuses on non-expert users, producing a graphical user interface for running reconstruction pipelines on acquired data and uploading image data and metadata to an XNAT database. Finally, Aim 3 focuses

on community awareness, accessibility, and documentation for the MCR Toolkit, ensuring the Toolkit is available and easy to use beyond the scope of this supplement and the parent grant. Successful completion of this work will transform the MCR Toolkit from a collection of tools useful to experts in the field to a platform for open science

among CT, basic science, and clinical researchers. Widespread adoption of the Toolkit is expected thanks to the unique benefits it provides for photon counting and dynamic CT which are not currently provided by open-source software alternatives.