ASL-BIDS, the brain imaging data structure extension for arterial spin labeling

Purpose/Introduction: The Brain Imaging Data Structure (BIDS) is a recently developed data storage standard, that meets the need for a structured manner to organize imaging data in the age of big datasets and data sharing (https://bids.neuroimaging.io). 1 This abstract presents a BIDS extension for ASL, which only supports ASL approaches as recommended in the ASL acquisition consensus paper, and several M0 calibration approaches. 2

Subjects and Methods: A group of ASL experts initiated this extension by defining several concepts and preparing a first draft. This draft was shared online from May 2017 until March 2019 with the international ASL community, and several teleconference and face-to-face meetings were organised. Per BIDS convention, existing BIDS fields were reused for the ASL-BIDS extension if possible. The BIDS fields names were based on the NEMA ASL DICOM fields, where possible. Additionally, three example datasets were collected 3
and efforts were initiated to adapt existing ASL analysis tools and the BIDS validator for ASL-BIDS compatibility.

Results: Six concepts were defined to allow a uniform yet flexible ASL-BIDS specification. First, it was decided to focus solely on the implementation of the ASL approaches discussed in the ASL consensus paper: single- and multi-delay, pulsed, continuous, and pseudo-continuous ASL. 5 Second, the BIDS-structure consists of two mandatory files and several optional files (Fig. 1). Third, it is obligatory to keep the ASL time series in the original acquisition order in a 4D NIfTI file, including any M0, if it was part of the original ASL
time series. If an M0 image was acquired separately, it should be stored as a separate NIfTI file. The ASL-context BIDS field explains the content of each volume in the ASL time series. Fourth, the derivative images DeltaM and CBF are considered to be raw images if the ASL-sequence or vendor only provided derivative images, lacking raw data. This principle follows the prioritization shown in Fig. 2. Fifth, all ASL data need to be stored in at least 32 bit floating point, without any scale slopes. Some vendor implementations store scaled ASL data to increase the precision of the stored data within the traditional 12 bit DICOM files. It is the responsibility of the DICOM to BIDS conversion to apply any existing scale slopes. Sixth, it is recommended to specify as much information as labeling as possible: the exact location of the labeling plane and the labeling efficiency.

Discussion/Conclusion: The current ASL-BIDS extension is restricted to the ASL approaches recommended by the consensus paper. 1 With the current development of more advanced ASL approaches, such as time-encoded and velocity-selective ASL, the ASL-BIDS may be extended for these technique. Also, a derivatives extension for ASL is anticipated.