Background Medical imaging plays a major role in the diagnosis and treatment of cardiovascular disease. The process of generating a medical image consists of two parts; data acquisition and image reconstruction. Image reconstruction transforms the acquired raw data signals into images that can be interpreted by clinician to aid in the diagnosis of a disease or used to guide a procedure. The raw data is frequently corrupted by instrument imperfections and patient motion. It is also common for datasets to be incomplete since there is a limited amount of time or other patient exposure available for data acquisition. Consequently, modern image reconstruction software is fairly complex. The source code for these complex image reconstruction algorithms is most often proprietary information retained by the system vendors and scientists working in the field of medical image reconstructions are forced to implement their own versions of existing algorithms that they then build on and improve. Because of this reimplementation and lack of open standards, many of the published literature on medical image reconstruction is not reproducible. The overarching goal of this project is to develop novel advanced image reconstruction algorithms and to do that in such a way that other scientists (and system vendors) can reproduce the presented results and use the methods in future work. The Laboratory of Imaging Technology, NHLBI, is particularly focused on Magnetic Resonance Imaging (MRI) techniques, but the developed principles apply to other techniques as well. Goals/Objectives The Laboratory of Imaging Technology develops and maintains two major software packages that support ongoing research projects. The first is the ISMRM Raw Data format (https://ismrmrd.github.io), which is an open raw data standard for MR experiments. It is a requirement for sharing algorithms and methods that there is common understanding of how to describe the raw data and this package provides a framework for this. We also aim to maintain data conversion tools from major device manufacturers proprietary formats to this vendor-independent format. The second software package is the Gadgetron (https://gadgetron.github.io), which is an advanced image reconstruction package that contains toolboxes and a streaming pipeline architecture for processing the raw data that is acquired by the imaging instrument.
We aim to expand this software package and support the growing user base around the world. There are a number of technical innovations that we are currently pursuing: - Expansion of the ISMRMRD format to include waveforms and telemetry from other instruments - Formal definition and implementation of ISMRMRD communication protocol - The use of cloud computing for MRI reconstruction - MRI raw data compression - Correction of measurement system imperfections - Tight integration of the Gadgetron with specific vendor instruments In addition to these infrastructure goals, we are developing and testing new imaging reconstruction techniques to solve specific clinical questions: - Real-time imaging sequences for interventional MRI - Real-time measurements of blood flow - Motion corrected, free-breathing techniques for measuring cardiac function and parametric maps. - Quantitative assessment of myocardial perfusion Progress in fiscal year 2018 In the past year, significant progress has been made on the integration of machine learning into the Gadgetron framework. This is motivated by the increasing demand for machine learning algorithms in image reconstruction and processing. A framework for network training and application has been integrated into the Gadgetron infrastructure and tested for processing tasks. This integration will enable fast prototyping of new algorithms and easy deployment in the clinical research environment. As we move to non-Cartesian imaging for an increasing number of applications, we have initiated the development of a generic reconstruction pipeline for non-Cartesian imaging. This pipeline is intended to be flexible such that a variety of acquisition types can be managed by the infrastructure. This includes the correction of system imperfections and management of physiological information for functional scans. The next generation of the ISMRM raw data format has been developed and tested. This new format extends the definition of the data format and enables simultaneous streaming of MRI raw data with other relevant waveforms, including physiological and system data. Access to these data will expand the capabilities of our reconstruction platform, the Gadgetron. Our cloud deployments of the Gadgetron have been frequently updated to include new features. The cloud based reconstruction platform is used by our collaborators for clinical applications that demand high-performance computing. We continue to work with the MRI community more broadly on both the Gadgetron and ISMRM Raw Data software packages.
Showing the most recent 10 out of 11 publications
