This section of the course provides an overview of the processing requirements required by emerging MR spectroscopic imaging approaches and describes a roadmap towards their automation and adoption within a clinical context.
Part I – An overview of data scope and basic SI processing workflow, followed by a discussion of workflow extension and added complexity due to cutting edge changes in SI acquisition.
Whole brain SI data sets can have huge memory management requirements. Even with undersampled and sparse data acquisition methods, multi-channel receive coils and reconstruction algorithms may result in dozens or more copies of the full-size SI matrix within the workflow. Also, while processing array requirements may be temporary, the final result of spectral fitting and analysis algorithms can yield multiple gigabyte arrays of results which need to be stored and accessed for statistical analysis or data provenance. CPU demands for cutting-edge SI processing algorithms are also high, if for no other reason than the amounts of data that must be processed. However, each methodology used to reduce or speed up data acquisition imposes additional data processing steps and complexity. Finally, until some consensus on SI methodology is arrived at, few of these processing and spectral analysis methods have undergone significant optimization, as many are still under development or being tweaked in response to new acquisition methods. Despite significant increases in computing hardware in the past decades, SI processing workflows still stretch and exceed typical computational environments, both in data management and algorithmic complexity that is not trivially parallelizable.
Part II – A brief discussion of the role of spectral fitting and analysis in addressing specific clinical and clinical research needs.
Basic research for SI has provided us with an array of tools for measuring a variety of metabolites. However, the ultimate goal for clinical SI development is to provide high quality and robust measures of metabolites that are of use to clinicians. This presumes that the tools will be used in a protocol designed to meet a concise description of desired metabolites, whether in a cohort or longitudinal (or both) context, data quantitation and/or reduction to address a specific question. Any given protocol might include omitting or reduction of spectral quality in certain metabolites to ensure quality in those that are required. Most SI tools have been developed for the more general purpose of acquiring as many metabolites across as large an anatomic volume as is possible, without specific thought as to a use case. We will show examples of existing third party solutions for SI measures in clinical settings.
Part III – A discussion of both the opportunities and barriers within a roadmap for establishing comprehensive SI workflows in the clinic.
The creation of SI workflows as standard products on manufacturers platforms include a number of barriers, but also some recent opportunities. 1) There is a need for fully automated processing, spectral analysis and metabolic reporting to enable SI measurements to be accessed at locations without MR physics support. 2) MR data taken with manufacturer products are required to conform to DICOM compliant data management and reporting standards. Vendor support is growing for the existing DICOM MR Spectroscopy Data Object and DICOM Structured Report standards. 3) Timely processing of cutting-edge SI methods that acquire data within times that are acceptable for patient will require a variety of computational hardware improvements, including memory, CPU and GPU, either within MR platforms or through cloud processing, and 4) SI usage within a clinical context will require a demand for specific clinical ‘killer apps’ that drive the development of these techniques towards WIP and eventually manufacturer products. We will present existing solutions that have been come about by overcoming, or in spite of, these limitations.