A Beginner's Guide to Arterial Spin Labeling (ASL) Image Processing

Abstract

Arterial spin labeling (ASL) is a non-invasive and cost-effective MRI technique for brain perfusion measurements. While it has developed into a robust technique for scientific and clinical use, its image processing can still be daunting. The 2019 Ann Arbor ISMRM ASL working group established that education is one of the main areas that can accelerate the use of ASL in research and clinical practice. Specifically, the post-acquisition processing of ASL images and their preparation for region-of-interest or voxel-wise statistical analyses is a topic that has not yet received much educational attention. This educational review is aimed at those with an interest in ASL image processing and analysis. We provide summaries of all typical ASL processing steps on both single-subject and group levels. The readers are assumed to have a basic understanding of cerebral perfusion (patho) physiology; a basic level of programming or image analysis is not required. Starting with an introduction of the physiology and MRI technique behind ASL, and how they interact with the image processing, we present an overview of processing pipelines and explain the specific ASL processing steps. Example video and image illustrations of ASL studies of different cases, as well as model calculations, help the reader develop an understanding of which processing steps to check for their own analyses. Some of the educational content can be extrapolated to the processing of other MRI data. We anticipate that this educational review will help accelerate the application of ASL MRI for clinical brain research.

Keywords: MRI; arterial spin labeling; cerebral blood flow; graphical user interface; image processing; perfusion; processing pipeline.

Figure 1

Overview of the PASL, CASL, and PCASL techniques The red H circles in the schematic overview images represent water protons in the blood that get labeled — depicted as red X marks — by a radiofrequency pulse (flash) in the labeling plane in the internal carotid arteries. Other extracranial arteries are labeled too but are not depicted here. Advantages and drawbacks are in comparison between the techniques. LD: labeling duration; RF: radiofrequency; TR: repetition time.

Figure 2

Example transverse slices of intermediate and final ASL images as quality control provided by ExploreASL, from a single healthy volunteer (female, 79 years of age) with 2D EPI PCASL sequence. (A) T1-weighted structural image in standard space; (B) T2 FLAIR image (pre-processed); (C) M0 image; (D) an average raw non-subtracted ASL control image; (E) the final CBF image; (F) a sagittal 2D time-of-flight vessel scout used to position the labeling plane.

Figure 3

Overview of all ASL processing steps on single-subject and group-level. The first stage, data conversion and sharing, prepares data for the actual image processing. The second and third stages are structural data processing and ASL data processing (all single-subject level), respectively. The fourth stage concerns group-level processing. CBF, cerebral blood flow; PVC, partial volume correction; ROI, region of interest [adapted from (8)].

Figure 4

Principle of white matter lesion correction. A T2 FLAIR image (A) with lesion hyperintensities, corresponding to an original T1-weighted image; (B) and a WM segmentation image; (C) from a 70-year-old female. The lesions are automatically identified and segmented on the T2 FLAIR; (D) which are used as a mask to define hypointense regions in T1-weighted images that could be misinterpreted as GM. Based on these lesion masks, the original T1-weighted image is corrected; (E) and shown with the final WM segmentation overlaid in red (F).

Figure 5

The effect of normalization on T1-weighted images. Left column: original images before normalization (native space), right column: after normalization to MNI space. Note how size differences and atrophy are largely compensated.

Figure 6

The effect of motion correction. (A) Perfusion image without motion correction; (B) The same perfusion image after motion correction. Data were obtained with pseudo-continuous labeling and 2D EPI readout without background suppression on a Siemens scanner. The original data had a single yaw rotation (subject turning the head from left to right). Motion artifacts can be easily recognized by prominent patterns of very high and very low (i.e., negative) signals next to each other in areas where the static tissue has high contrast differences — visible as rims around the brain edge and skull.