Myocardial tagging by cardiovascular magnetic resonance: evolution of techniques--pulse sequences, analysis algorithms, and applications

Abstract

Cardiovascular magnetic resonance (CMR) tagging has been established as an essential technique for measuring regional myocardial function. It allows quantification of local intramyocardial motion measures, e.g. strain and strain rate. The invention of CMR tagging came in the late eighties, where the technique allowed for the first time for visualizing transmural myocardial movement without having to implant physical markers. This new idea opened the door for a series of developments and improvements that continue up to the present time. Different tagging techniques are currently available that are more extensive, improved, and sophisticated than they were twenty years ago. Each of these techniques has different versions for improved resolution, signal-to-noise ratio (SNR), scan time, anatomical coverage, three-dimensional capability, and image quality. The tagging techniques covered in this article can be broadly divided into two main categories: 1) Basic techniques, which include magnetization saturation, spatial modulation of magnetization (SPAMM), delay alternating with nutations for tailored excitation (DANTE), and complementary SPAMM (CSPAMM); and 2) Advanced techniques, which include harmonic phase (HARP), displacement encoding with stimulated echoes (DENSE), and strain encoding (SENC). Although most of these techniques were developed by separate groups and evolved from different backgrounds, they are in fact closely related to each other, and they can be interpreted from more than one perspective. Some of these techniques even followed parallel paths of developments, as illustrated in the article. As each technique has its own advantages, some efforts have been made to combine different techniques together for improved image quality or composite information acquisition. In this review, different developments in pulse sequences and related image processing techniques are described along with the necessities that led to their invention, which makes this article easy to read and the covered techniques easy to follow. Major studies that applied CMR tagging for studying myocardial mechanics are also summarized. Finally, the current article includes a plethora of ideas and techniques with over 300 references that motivate the reader to think about the future of CMR tagging.

Figure 1

Tagging by magnetization saturation. The technique is based on implementing slice-selective magnetization saturated planes orthogonal to the imaging slice. (a) Pulse sequence. Each tagging plane needs a slice-selective RF pulse during the tagging stage. The imaging part follows tagging, and the figure shows conventional Cartesian acquisition (RO = readout, PE = phase encoding, SS = slice selection). (b) Tagging planes and imaging slice.

Figure 2

SPAMM tagging. (a) SPAMM pulse sequence. The tagging part consists of only two non-selective RF pulses (usually, 90° each), separated by the tagging gradient in the tagging direction, and followed by a large crusher gradient. The imaging part shows conventional Cartesian k-space acquisition (RO = readout, PE = phase encoding, SS = slice selection). This sequence creates parallel tag lines orthogonal to the x-axis. (b) Example of a SPAMM grid-tagged image showing left ventricle (LV) and right ventricle (RV). Note that this grid pattern needs the application an extra tagging stage (in the orthogonal direction) next to the first one before imaging takes place. Note also that the dark myocardium between the tag lines is not completely black due to longitudinal relaxation. (c) Illustration of spins evolution during different time points in the tagging stage, as follows: immediately before tagging application (time point 1), the magnetization (M) is at equilibrium state in the longitudinal direction. Immediately after the application of the first RF pulse (time point 2), the magnetization is tipped into the transverse direction by certain flip angle (45° RF pulses are assumed here for illustration). The tagging gradient then follows, which disperses the spins in the tagging direction (x-direction in this case), such that by the end of the gradient pulse (time point 3), the spins are modulated by incremental phase shifts along the x-axis (the figure shows all vectors emerging from the origin just for simplicity). The second tagging RF pulse tips the resulting modulated magnetization by another 45° into the transverse direction to result in spins modulated as shown at time point (4). A crusher gradient immediately follows to eliminate transverse magnetization components, leaving only the longitudinal parts, which show a sinusoidal pattern along the x-axis with values ranging from 0 to M.

Figure 3

DANTE pulse sequence. The tagging stage contains a series of hard (non-selective) RF pulses run simultaneously with accompanying gradient in the tagging direction (1-D tagging is shown here). The imaging stage follows after tagging. The figure shows conventional Cartesian k-space acquisition (RO = readout, PE = phase encoding, SS = slice following).

Figure 4

CSPAMM tagging. (a) CSPAMM pulse sequence. The sequence runs two SPAMM sequences, with the polarity of the second tagging RF pulse changed in the second SPAMM acquisition. Notice also the ramped flip angles of the imaging RF pulses to compensate for fading tagging. (b) Example of a CSPAMM grid-tagged image. Notice that non-tagged tissues appear black due to the elimination of the offset DC signal. (c) The concept of magnetization subtraction in CSPAMM. Two scans are acquired as shown in the pulse sequence, which results in positive and negative sinusoidal tagging patterns from the first and second scans, respectively. With time, the tagging patterns experience longitudinal relaxation, trying to reach equilibrium (M₀). The relaxation has two effects on the tagging pattern: the peak-to-peak (AC) magnitude is decreased; and the tagging pattern now has non-zero average (DC) value. However, the DC component is the same in both scans. Thus, at any time point, when the two acquired images are subtracted, the DC component cancels out and the peak-to peak magnitude doubles as shown.

Figure 5

Slice following. (a) Pulse sequence. The pulse sequence is similar to CSPAMM, except that one of the tagging RF pulses is replaced by a slice-selective one to create a thin tagged slice. During imaging, a thicker slice is excited. (b) A thin tagged slice is prescribed, which experiences deformation and displacement by the imaging time. A thick imaging slice is selected to accommodate the tagged slice displacement.

Figure 6

bSSFP pulse sequence. The tagging part is similar to regular tagging. However, the imaging part is different, where the gradients are balanced in all three axes. Zero net gradient is achieved during each repetition time (TR) to reduce magnetization dephasing and enhance acquired signal.

Figure 7

Tagging by consecutive application of 1-D tagging in orthogonal axes. (a) 2-D case. Instead of applying tagging in two orthogonal axes in the same scan (to create a grid tagged pattern), two 1-D tagging scans are conducted consecutively, so that only a small portion of k-space that includes the signal peaks is acquired each time, instead of acquiring the whole k-space in the first case. (b) Extension to 3-D case. Instead of applying tagging in three orthogonal directions in the same scan, three consecutive 1-D tagging scans are conducted consecutively to significantly save data acquisition time. (c) The three orthogonal 1-D tagged images. Short-axis slices are tagged in both horizontal and vertical directions in two separate acquisitions, and four-chamber images are tagged in the horizontal direction. The displacement information from all three sets of images are combined together to obtain 3-D displacement information.

Figure 8

Heart 3-D modeling. The spline lines (or planes) are fitted to the tag planes, and their intersections create marker points that are used to build a 3-D finite-element model of the heart. The model is used for displacement calculations.

Figure 9

HARP tagging. (a) Original SPAMM tagged image. (b) K-space of the tagged image. HARP applies a spatial bandpass filter to extract only the first harmonic peak. (c) Magnitude HARP image is created by Fourier transforming the modified k-space and obtaining the magnitude data. (d) HARP phase image is created by Fourier transforming the modified k-space and obtaining the phase data. (e) Multiplying the HARP magnitude and phase images results in an image with modulation pattern very similar to the original tagged image. However, the pattern here represents the wrapped tissue phase. Myocardium displacement tracking can be conducted by simply tracking the tissue phase from frame to frame. (f) An example of a grid-tagged image analyzed with HARP, which shows myocardial circumferential strain.

Figure 10

3D-HARP. (a) Two series of tagged images are acquired: 1) a stack of parallel short-axis grid tagged images to obtain in-plane x-y displacements; and 2) a radial set of long-axis line-tagged images to obtain through-plane (in the z-direction) displacements. (b) The HARP concept is used to track myocardium. The intersection of tag line with the image plane (point P1) is linked to the nearest point on the next frame with the same characteristic phase (point P2).

Figure 11

zHARP pulse sequence. The pulse sequence is derived from CSPAMM. Two CSPAMM images are acquired with horizontal and vertical lines to obtain in-plane strains. Through-plane strain is obtained by adding positive (light gray) and negative (dark gray) z-gradients after the slice selection pulse of the horizontal and vertical tagged images, respectively.

Figure 12

DENSE. (a) DENSE pulse sequence is derived from STEAM sequence. Displacement encoding (modulation) and decoding (demodulation) gradients are added after the first and third RF pulses, respectively. Both gradients have equal magnitudes, and they are applied in the direction where displacement information to be obtained. During the mixing period in-between modulation and demodulation, large crusher gradients are applied to dephase any transverse magnetization, such that the remaining magnetization is stored in the longitudinal direction, where it experiences only longitudinal relaxation. The time between the first and second RF pulses is equal to the time between the third RF pulse and data acquisition (= echo time (TE)/2). TM = mixing time; RF = radiofrequency; RO = readout; PE = phase encoding; SS = slice selection. (b) Example of DENSE image of the left ventricle, where representative vectors are drawn at each pixel, with the vector magnitude and direction represent the displacement value and orientation, respectively.

Figure 13

DENSE k-space. Three echoes are generated in DENSE: stimulated echo, stimulated anti-echo, and T1 relaxation echo. Only the stimulated echo contains the displacement information of interest, while other echoes are to be suppressed. The bigger the displacement encoding frequency, the larger the separation between the stimulated echo and the T1 relaxation echo, which results in displacement-encoded image not corrupted by other signals. The T1 relaxation echo can be also suppressed by inversion recovery or phase cycling (as in CSPAMM). The anti-echo may be added to the stimulated echo after phase correction to improve SNR.

Figure 14

SENC. (a) SENC idea. SENC applies tagging as in conventional SPAMM. However, the tag planes are applied parallel to and inside the imaging slice, thus through-plane strain is recorded. (b) SENC pulse sequence. The tagging part is similar to that in conventional SENC, except that the modulating gradient is applied in the through-plane (z-) direction. During the imaging stage, a tuning gradient is applied in the z-direction after slice selection and before data acquisition. By acquiring two images with different tunings, strain information can be decoded. (c) Example of strain images. Low-tuning (LT) image shows non-contracting tissues; high-tuning (HT) image shows contracting myocardium. The LT and HT images are combined to show color-coded strain map. Note that this four-chamber slice shows circumferential strain, and note the big non-contacting apical region.

Figure 15

fast-SENC and sf-fast-SENC. (a) fast-SENC pulse sequence. The tagging part has two slice-selective RF pulses applied in the x- and y-directions, respectively, to restrict tagging to a small square region, which allows for reducing FOV without much compromising spatial resolution. The second modification in fast-SENC is the implementation of alternating LT and HT tunings in consecutive heart phases to reduce scan time in half. The third modification is the application of spiral acquisition to reduce the imaging time. sf-fast-SENC pulse sequence (not shown) is similar to fast-SENC, except that it replaces the first tagging RF pulse by 2-D cylindrical excitation in the x-y direction and the second RF pulse by a slice-selective pulse in the z-direction to restrict the tagged region to a small disc-shaped region of the intersection of the two excited regions. (b) Shapes showing the tagged and imaged regions in SENC, fast-SENC, and sf-fast-SENC. In SENC, all the body is tagged; thus large FOV imaging is required. In fast-SENC, tagging is restricted to the intersection of the two orthogonal slabs. This allows for using small imaging FOV. In sf-fast-SENC, tagging is restricted to a disc-shaped thin region, which is the intersection of the 2-D cylindrical region and the orthogonal slice. This allows for applying slice following by imaging with a thicker slice that has reduced FOV. (c) In SENC, two scans are required to capture LT and HT images at different heart phases. In fast-SENC and sf-fast-SENC, the tunings are applied alternatively in one scan only.

Figure 16

C-SENC. (a) C-SENC pulse sequence. The sequence is similar to SENC with the exception that an additional image is acquired without any tuning (no-tuning, NT). (b) Different signals acquired in C-SENC. The LT and HT signals result from non-contracting and contracting tissues, respectively. At zero frequency, another signal peak arises from relaxed magnetization. Infarcted myocardial results in larger signal than normal myocardium 15 minutes after Gd injection due to its preferential accumulation in infarction. (c) Example of C-SENC images. The NT image shows noticeable difference between normal and infarcted myocardium. The C-SENC image is acquired by combining functional information from LT and HT images with viability information from NT image.