Automatic analysis of tagged images of laminar fluid flow

Abstract

The visualization and measurement of motion using the distortion of magnetization “tags” in magnetic resonance imaging was first developed for cardiac imaging ( l-3)) but is now beginning to be applied to other, nonmedical problems (4, 5). Several techniques for generat ing tags are available, all of which rely on saturation of the magnetization to give a low-intensity grid line. Originally, tags were generated using a binomial RF pulse (3)) although, more recently, DANTE pulses (6) have also been emp loyed. Recent progress has been made in analyzing the tag deformation for components of strain ( 7), but the identification of unambiguous points in the tagging pattern has in general been by hand (8, 9). We present here the results of mod ifying the tagging pattern to give images that are more readily analyzable by computer. The original motivation was to allow the analysis of 3D MR images, which is an extremely difficult manual process; with that achieved the obvious consequence is that 2D images can also be handled with increased accuracy, efficiency, and consistency. The phi losophy for these tagging experiments is slightly different from that normally emp loyed. Consider the behavior of a single element of a tag line in a fluid as the fluid moves; al though it may increase in length, due to fluid motion, it is impossible to tell where the extension occurred along the line element. Hence, only the displacement component perpendicular to the tag line element can be determined, the only points that can be unambiguously traced as they move are the tag crossing points. However, identification of these intersections can prove difficult since, as the geometry of the grid changes, the crossings are no longer perpendicular and are harder to identify by a simple matched filtering and in general must be identified manually, al though some studies have opted for a tag-following algorithm ( 7). As we now demonstrate, if the tag lines are widened the unsaturated areas become spots, which, in the lim it of very wide tag lines, unambiguously mark points in the fluid. Bright spots are much easier to identify automatically and locate accurately in an image by simple thresholding than line crossings, with the obvious consequences for computer-assisted analysis of strain and displacement. There are several approaches to generat ing wide saturation bands in an image, all of which can be derived approximately from Fourier analysis of the saturation pattern required. If the tag profile is to be square, then a sine modu lation must be applied to the flip angles of the pulses that make up the DANTE excitation. The sine-modulated DANTE (SINC-DANTE) approach was emp loyed recently to generate 50% tags pre-