Introduction

The Open Science Initiative for Perfusion Imaging (OSIPI) (https://www.osipi.org/) is an international collaborative effort, under the guidance of the ISMRM perfusion study group, with the mission to promote the sharing of MR perfusion imaging software. The goal of OSIPI is to reduce duplicated development and improve the reproducibility of perfusion imaging research. Arterial spin labeling (ASL) is a non-invasive MR perfusion imaging technique that has been well-recognized in research and has considerable potential in the clinic1,2. Task Force 2.2 of OSIPI (https://www.osipi.org/task-force-2-2/) aims to collect and organize current libraries and functionalities for ASL data processing. This is specifically targeted for intermediate and advanced developers of ASL perfusion methods, who are looking for specific functionalities and/or development templates, without the need to develop the code from scratch. This abstract provides an overview of this Task Force including the milestones and the current status.

Methods

The Task Force 2.2 includes team members from eight different institutions across the globe, including the US, the Netherlands, UK, China, and Germany. The following five milestones have been established to successfully achieve the goals of this task force (Figure 1).

1. Milestone #1 was to define the scope of the ASL code library. It determined the functionalities to be targeted, identified specific components and the corresponding timelines, identified, and recruited specific leads, and identified the components that need interactions with other task forces.
2. Milestone #2 was to gather existing ASL image processing code snippets. It consisted in reaching out to the teams/members to gather the code(s) and create a repository. Furthermore, we identified code(s) for various major vendors, documented the code origin, functionality, and sample data from the owner.
3. Milestone #3 is to publish the collected ASL library version 1.0.
4. Milestone #4 is to finalize the ASL processing code and libraries. It will develop unit tests for code contributions in the library, execute tests and collect results and feedback to developers, and reiterate the processes in Milestone #4 to finalize the code(s) and libraries if necessary.
5. Milestone #5 is to complete documentation and to publish the final library.

We have collected functions and scripts that can enable the individual functionalities of the ASL pipelines. The input data are restricted to images (and not k-space) for this initial phase. Commonly-used functionalities for ASL brain perfusion analysis have been collected, including motion correction, brain extraction/skull stripping, segmentation of high-resolution images, co-registration of ASL images to high-resolution T1-weighted, T2-weighted, and/or FLAIR images, proton-density (M0) processing, cerebral blood flow (CBF) quantification model, arterial transit time quantification, time/Hadamard encoding and decoding, nuisance correction to remove covariates from ASL time series, outlier removal or denoising, quality control, normalization to common template (e.g. MNI space), partial volume correction, and deep learning-based methods. The most used software programming languages were considered. The focus was primarily focused on Python functions and scripts, given its increasing popularity and our emphasis on open-source languages. More specifically, detailed Python packages’ versions were expected. Submission with a docker image was also encouraged. However, given the substantial existing codebase in Matlab, Bash, and C++, these languages were also considered. Additional languages such as R have been also taken into account.

Results

The scope definition (Milestone #1) was successfully completed in August 2020. The current library focuses on brain applications and images of vendors’ products. It includes the pulsed ASL (PASL), continuous ASL (CASL), and pseudo continuous ASL (pCASL) processing for brain imaging. The initial call for code collection (Milestone #2) has been sent out to OSIPI, ISMRM Perfusion study group, SPM mailing list, and FSL mailing list in September 2020. We also reached out to specific contributors who expressed interest in ASL pipelines. It included 14 different groups and seven positive responses were received.
We received contributions from seven different groups, including the ASLtbx and DL-ASL from the University of Maryland School of Medicine, the ASL-Angio reader and super-selective pCASL CBF from the Universitätsklinikum Hamburg-Eppendorf, the QuBIc groups from the Oxford Institute of Biomedical Engineering, ASL-MRICloud from the Johns Hopkins University School of Medicine, Basil from the University of Nottingham, and Clinical ASL from UMC Utrecht. The source code is under review by our team and the functionalities are being separated and documented. The source scripts are organized in Github, and on Google Drive. In addition, an open-source repository will host and share all the codes.

Discussion

The successful completion of this project will provide a platform for intermediate and advanced developers of ASL perfusion methods to exchange code for image analysis in the future. The scope of this first 2-year project is focused on human brain applications and the initial data type will be focused on images provided by vendors. However, we are open to the option of extending this to other applications and image reconstruction in the future. Users who are interested in the project are welcome to contact us, attend our virtual meetings (3rd Wednesdays of each month from 9:30-10:30 EST. Google Meet: meet.google.com/idy-qinf-xar), contribute to our library (Google Drive: https://drive.google.com/drive/folders/16VPJ4Wq-YW8GA5cddMpKgWKs4Ua5dPEV) and explore the library (Github: https://github.com/OSIPI/OSIPI-ASL-toolbox).

Acknowledgements

HM is supported by the Dutch Heart Foundation (2020T049), and by the Eurostars-2 joint programme with co-funding from the European Union Horizon 2020 research and innovation programme, provided by the Netherlands Enterprise Agency (RvO).