Motivation: This proposal, titled Interoperable Software Platform for Reproducible Research and Clinical Translation of MRI, is in response to the U24 funding opportunity RFA-EB-18-002, Resources for Technology Dissemination. Magnetic resonance imaging (MRI) is non-invasive, non-ionizing, and offers superb soft tissue contrast, but is traditionally limited by long scan times. Recently, advances in numerical image reconstruction and availability of powerful hardware platforms have led to new MRI scanning techniques with dramatic reductions in scan times. However, the associated computational sophistication has posed a large barrier to reproducibil- ity and clinical translation. This proposal addresses this fundamental issue by establishing best practices and infrastructure for reproducible research in MRI. Initial work toward this goal spanning six years has led to the development of the BART software toolbox for computational MRI. BART implements advanced MRI reconstruction algorithms in an extensible manner so that new technological advances can build off of the collective progress in the ?eld. Supported computational back- ends including multi-CPU and multi-GPU architectures afford ef?cient use in a clinical translation environment. Project dissemination has been met with strong interest from the international MRI research community, having grown a user-base spanning over 50 academic and industry sites. Nonetheless, current limitations in project in- frastructure and support have hindered more widespread dissemination. Therefore, the major emphasis here is expanding development to improve usability, creation of written and audio-visual educational material, integration with other tools, cloud-based support, and software reliability. This will (1) provide new users common ground for starting new projects, (2) allow them to use their existing work?ows with BART, (3) move to more accessible computation platforms, and (4) reliably translate their work into clinical practice. Approach: The project will proceed with four interrelated aims, supported by user training activities.
Aim 1 will focus on adding comprehensive documentation and creating example-based tutorials.
Aim 2 will expand interop- erability with software platforms and vendor tools used by the MRI community.
Aim 3 will complete infrastructure and backends for cloud and parallel computing.
Aim 4 will improve software reliability and quality assurance. The work will be disseminated through online material, webinars and workshops. Signi?cance: This work will enable development, creation and reproducibility of modern state-of-the art MRI reconstruction methods that rely on highly specialized data processing approaches. MRI development will be streamlined as new methods build off of reliable infrastructure and existing work. Improved sustainability and reliability will enable rapid dissemination of new work into clinical evaluation and practice while signi?cantly reducing the technical burden normally associated with clinical translation.
Magnetic Resonance Imaging (MRI) is one of the most versatile imaging modalities in biomedical imaging and has become an indispensable tool in neuro-science, clinical and pre-clinical research, and clinical practice. Ad- vances in numerical image reconstruction that leverage the massive growth in computation power have enabled the development of advanced MRI applications which were not previously possible, but the complexity renders many published studies dif?cult to reproduce and translate to practice. This work aims to provide a software framework for implementing new techniques in a reproducible and shareable format while providing interoper- ability with existing tools used by the MRI research community, with the goal of enabling clinical translation.
