Arterial spin labeling (ASL) MRI is a noninvasive method to quantify cerebral perfusion. With recent standardization of acquisition protocols (through the ASL white paper¹) and strong interest from the vendors, this previously underachieving tool finally seems to be getting its chance to catch up with its peers, fMRI and DTI, which were developed around the same time as ASL (about two decades ago) yet have enjoyed much more successful acceptance and applications. Not surprisingly, downloadable analysis toolboxes^2,3 for ASL are also somewhat primitive compared those for fMRI and DTI. Furthermore, when it comes to downloadable toolboxes in neuroimage analysis, most of us may have, at some point, experienced the frustration of software compatibility, including operating system platforms (e.g. Windows, Linux, or OSX), 32-bit vs. 64-bit, image analysis software upgrades (e.g. differences between SPM2, SPM5, SPM12). These issues may appear daunting to many of our end users, i.e. radiologists, neurologists, psychiatrists, neuroscientists, who are exactly the crowd we want to advocate this technique to. In this work, we describe a cloud-computing-based tool for comprehensive analysis of ASL MRI (www.MriCloud.org). The main strengths of this tool are that it does not depend on local computer operating system, does not require downloading or installation of any software, places no constraints on the CPU or memory capacity of the user’s computer, and that the developer rather than the user will take the responsibility of software upgrade.

General concept of the “ASL in the MriCloud” tool

The proposed ASL pipeline with an option of the multi-atlas brain segmentation is highly CPU intensive. Instead of performing computational analysis of the ASL data on the user’s local computer, the tool places the computational and programming burden on our internal or publicly available computation resources such as Computational Anatomy Gateway(http://info.teragrid.org/web-apps/html/views/tggateways)(Figure 1). To start, the user only needs to upload the raw ASL data onto the server using a web browser and click the submit button. Then, a few minutes later, quantitative perfusion outcomes will be available for downloading. The user can then download the outcomes (detailed below) and conduct their study-specific statistical analysis.

Web interface

Our ASL tool can be found at https://braingps.mricloud.org/asl. The infrastructure of the cloud computing is based the work of Mori et al⁴. The user interface(Figure 2) allows the uploading of raw ASL image, M0 image and T1-based anatomical data if applicable. The T1 data contains MPRAGE image and multi-atlas segmentation data, which can be obtained from another cloud tool in the same MriCloud service platform. Availability of the T1 data allows the normalization of the individual image into MNI template space for voxel-based analysis of cerebral blood flow (CBF) maps. Alternatively, the segmentation data can be used for ROI-based analysis. The user is also asked to provide several basic imaging acquisition parameters (e.g. multislice or 3D) and assumptions for CBF quantification. If the ASL scan was acquired following recommendations of the ASL white paper¹, default values can simply be used.

ASL analysis pipeline

The ASL scripts on the cloud server were written in Matlab 2013 and SPM12. The analysis procedure(Figure 3) followed that suggested in the ASL white paper1. Specifically, after motion correction, the difference image (control-label) is calculated. Next, M0 is obtained either through the uploaded M0 image or is calculated by estimating equilibrium magnetization from the control images. Then the difference image and M0 are incorporated in a kinetic model to calculate CBF map in ml/100g/min. If T1 data are uploaded, the individual-space CBF map is further normalized to MNI space. Since the T1 data also contains parcellations of the brain, regional CBF values within each parcellation are calculated.

Figure 4 illustrates the resulted CBF maps generated by the cloud analysis from one healthy subject, suggesting that the cloud-based ASL tool could calculate CBF maps reliably under various input conditions. Figure 5 shows an example of ROI definitions used in the cloud-analysis. CBF value of each ROI was calculated accordingly. The downloadable output of the analysis is 1)CBF map in individual space; 2)CBF map in MNI space, 3)a text file including regional CBF values as well as their voxel count.We have demonstrated a cloud-based analysis tool for ASL data. With this tool, the user can obtain CBF maps and ROI-averaged CBF values without having to install any software on the local computer. The maintenance of software upgrades will be performed by the developer. The CBF analysis algorithm follows recommendations of the ASL white paper. This tool may prove valuable and timely in accommodating the recent surging interest in ASL MRI.