MRI with Sub-Millisecond Temporal Resolution over a Reduced Field of View

Abstract

Purpose:

To demonstrate an MRI pulse sequence – Sub-millisecond Periodic Event Encoded Dynamic Imaging with a reduced field of view (or rFOV-SPEEDI) – for decreasing the scan times while achieving sub-millisecond temporal resolution.

Methods:

rFOV-SPEEDI was based on a variation of SPEEDI – get-SPEEDI, which employed each echo in an echo-train to sample a distinct k-space raster by synchronizing with a cyclic event. This can produce a set of time-resolved images of the cyclic event with a temporal resolution determined by the echo spacing (typically < 1 ms). rFOV-SPEEDI incorporated a 2D RF pulse into get-SPEEDI to limit the FOV, leading to reduction in phase-encoding steps and subsequently decreased scan times without compromising the spatial resolution. Two experiments were performed at 3 Tesla to illustrate rFOV-SPEEDI’s capability of capturing fast-changing electric currents in a phantom and the rapid opening and closing of aortic valve in human subjects over reduced FOVs. The results were compared with those from full FOV get-SPEEDI.

Results:

In the first experiment, the rapidly varying currents (50–200 Hz) were successfully captured with a temporal resolution of 0.8 ms, and agreed well with the applied currents. In the second experiment, the rapid opening and closing processes of aortic valve were clearly visualized with a temporal resolution of 0.6 ms over a reduced FOV (12×12 cm²). In both experiments, the acquisition times of rFOV-SPEEDI were decreased by 33%−50% relative to full FOV get-SPEEDI acquisitions and the spatial resolution was maintained.

Conclusion:

Reducing the FOV is a viable approach to shortening the scan times in SPEEDI, which is expected to help stimulate SPEEDI applications for studying ultrafast, cyclic physiological and biophysical processes over a focal region.

1. Introduction

Achieving an adequate temporal resolution is important for studying dynamic processes such as cardiac and other physiologic movements across a range of time scales. Recently, MRI techniques capable of sub-millisecond temporal resolution began to emerge (1–4). One of the promising techniques that can be readily implemented in multiple sequences on virtually all scanners is Sub-millisecond Periodic Event Encoded Dynamic Imaging, or SPEEDI (3). In this technique, MRI acquisition is time-locked to a cyclic event so that the same time point of an FID (or spin echo) signal from multiple TRs can be grouped into one of the distinct k-space matrices to resolve the time evolution of the cyclic event. The temporal resolution of SPEEDI is determined by the dwell time, which can be as short as several micro-seconds. SPEEDI has been successfully used to visualize ultrafast, cyclic dynamic changes such as eddy currents (3). Due to the reliance on phase-encoding, however, the acquisition time of the original SPEEDI pulse sequence was relatively long. To reduce the scan times, a variation of SPEEDI, known as gradient-echo-train SPEEDI or get-SPEEDI, was recently developed (2,5).

The scan times can be further decreased by limiting the number of phase-encoding steps with a reduced field-of-view (rFOV). This approach is particularly applicable to situations where the region of interest is spatially limited within a larger object, such as the heart in the chest wall. A number of reduced FOV techniques have been proposed over the past several decades (6,7). Among these, the approach that employs a 2D RF pulse has gained popularity, as illustrated in a number of publications (8–13). The 2D RF pulse can excite an in-plane strip within a slice to restrict the FOV along the phase-encoding (PE) direction, preventing aliasing artifact that would otherwise occur when using a FOV smaller than the object.

The purpose of this study is to incorporate an rFOV technique based on a 2D RF pulse into get-SPEEDI, which we call rFOV-SPEEDI, to achieve additional scan time reduction. We demonstrate the technical feasibility through two examples: (i) capturing fast-changing electric currents in a phantom, and (ii) visualizing the dynamics of rapid opening and closing of human aortic valve in vivo.

2. Methods

2.1. rFOV-SPEEDI Sequence

rFOV-SPEEDI is built upon a get-SPEEDI pulse sequence reported recently (2,5). Similar to the original SPEEDI sequence utilizing FID acquisitions (3), get-SPEEDI synchronizes each TR with a cyclic event under study (Figure 1A). Unlike the original SPEEDI sequence, get-SPEEDI utilizes an echo train, in which N gradient echoes are distributed across N separate k-space matrices, each corresponding to a distinct time point. The echo-train acquisition is repeated M times, each with a unique phase-encoding value, to sample full or partial k-space. The N resultant k-space matrices are individually reconstructed to produce a series of time-resolved images. The temporal resolution is determined by the echo spacing (esp), which is typically shorter than 1 ms. Compared to SPEEDI based on FID, get-SPEEDI can substantially shorten the acquisition times by eliminating phase-encoding in one dimension.

Figure 1:

An rFOV-SPEEDI pulse sequence (A) using a 2D RF pulse (B). Each TR is time-locked to a periodic event under investigation to acquire a train of gradient echoes. Each echo in the echo-train is assigned to a distinct k-space matrix (k-space 1, k-space 2, … k-space N). M in the figure represents the total number of TRs or the total number of phase-encoding steps required to sample k-space. After repeating the sequence M times, N k-space matrices, each containing M lines, can be obtained and reconstructed into N time-resolved images with a temporal resolution determined by the echo spacing (esp). In (A), the colors in the gradient lobes (or the gradient echoes) correspond to the colors of k-space matrices. In (B), the 2D RF pulse consists of 11 sub-pulses (e.g., each with a duration of 0.8 ms and a gap of 0.6 ms) that are played out during the odd gradient lobes in a “fly-back” transmit k-space design. The overall duration of the 2D RF pulse was 14.8 ms. G_RO: read-out gradient; G_PE: phase-encoding gradient; G_SS: slice-selection gradient.

In rFOV-SPEEDI (Figure 1A), the acquisition times were further reduced by limiting the FOV and accordingly decreasing the number of required phase-encoding steps (or M). Reduction in FOV was achieved by replacing the conventional excitation RF pulse in get-SPEEDI with a custom 2D RF pulse (Figure 1B), which was designed according to the descriptions in (14,15). The RF pulse consisted of eleven linear-phase sub-pulses that were designed using a Shinnar-Le-Roux (SLR) algorithm with a time-bandwidth product (TBP) of 3.2. The eleven sub-pulses were modulated by an SLR envelope pulse with a TBP of 3.3, a pulse width of 14.8 ms, and a linear phase. The resultant 2D RF pulse is illustrated in Figure 1B. In the spatial response of the 2D RF pulse, the gap between the two adjacent excitation bands was designed to adequately cover the imaged object in the phase-encoding direction to prevent aliasing artifacts from the side bands. The gradients accompanying the 2D RF pulse were designed by employing an echo-planar excitation k-space trajectory with the bi-polar gradient and the blip gradient directions assigned to the slice-selection and the phase-encoding directions, respectively (Figure 1B). A “fly-back” strategy was used by playing out the RF sub-pulses only during the odd gradient lobes to improve robustness against eddy currents. To shorten the entire 2D RF pulse width, the duration of the fly-back lobes was minimized by utilizing the maximal slew rate allowed by the Reilly’s curve (16). To enable 2D multi-slice imaging, a tilted transmit k-space trajectory was used (6), as detailed in (8,12).

2.2. Experiments

The rFOV-SPEEDI sequence in Figure 1 was implemented on a 3T GE MR750 scanner (General Electric Healthcare, Waukesha, Wisconsin). Two experimental studies were carried out to demonstrate the ability of rFOV-SPEEDI for reducing the scan times while maintaining spatial resolution over a limited FOV. The first experimental demonstration was conducted on a phantom that contained a rectangular wire loop submerged in a water bottle as described in a previous study (3). Two periodic current waveforms were delivered to the wire loop using a pulse generator Pulsepal (Sanworks, Stony Brook, NY) (17). The first current waveform, which imitated an action potential, consisted of two identical pulses, each with a 3 mA peak current and a duration of 5 ms, separated by an interval of 15 ms (i.e., 50 Hz). The second current waveform contained a series of rectangular pulses (2 ms ON and 3 ms OFF; 200 Hz). Synchronization between the current waveforms and the rFOV-SPEEDI sequence was accomplished by using the RF unblanking signal from the scanner as a trigger. The echo-train acquisition spanned a time window of 38 ms, covering the duration of the time-varying electric current. For comparison, the experiment was performed with both full FOV using get-SPEEDI (FOV = 12 cm × 12 cm, matrix = 64 × 64, pixel size = 1.875 mm × 1.875 mm, esp = 0.6 ms, and scan time = 2:08) and reduced FOV using rFOV-SPEEDI (FOV = 4 cm × 4 cm, matrix = 32 × 32, pixel size = 1.25 mm × 1.25 mm, esp = 0.8 ms, and scan time = 1:04). The other parameters common to both sequences were: TR = 2000 ms, TE = 3.8 ms, and slice thickness = 5 mm. For each sequence, two scans were performed with and without current in the wire loop, respectively, followed by a phase difference calculation to remove phase errors caused by unwanted B₀-field inhomogeneities.

The second experiment was designed to visualize the rapid opening and closing of the human aortic valve. With approval from the Institutional Review Board, ECG-gated cardiac images were acquired from two healthy subjects with a 32-channel phased-array cardiac coil. To visualize the entire opening and closing processes of the aortic valve, an “interleaved multi-phase” acquisition scheme was employed as shown in Figure 2. With a first trigger delay (delay1 in Figure 2), a total of L acquisition blocks was obtained to cover multiple cardiac phases, each acquisition block producing a total of N time-resolved k-space matrices after M TRs. To fill the temporal gap in-between the acquisition blocks due to the pulse sequence components other than the actual acquisition windows, a second trigger delay (delay2) was applied to obtain another L acquisition blocks. If the temporal gap remains, additional trigger delays can be added to provide a complete characterization of the entire aortic valve movement.

Figure 2:

A schematic of the interleaved multi-phase acquisition strategy used in the aortic valve imaging experiment. The red downward arrows indicate ECG trigger that was used to synchronize with the SPEEDI sequences. With a first trigger delay (delay1; blue), L acquisition blocks (each contains N images) were obtained within an R-R interval to cover a specific time span of the dynamic process (blue). To fill the temporal gap between the acquisition blocks, another set of L acquisition blocks was obtained with a second trigger delay (delay2; red). If the temporal gap remains, additional trigger delays can be added to achieve a complete coverage of the whole dynamic process. In doing so, a continuous characterization of the entire aortic valve movement can be achieved with a total of (number of trigger delays) × L × N images.

The above interleaved multi-phase acquisition strategy was applied to both rFOV-SPEEDI and full FOV get-SPEEDI. In the rFOV-SPEEDI acquisition the following parameters were used: TR = 29 ms, TE = 8.8 ms, flip angle = 10°, slice thickness = 8 mm, FOV = 12 cm × 12 cm, reconstruction matrix size = 60 × 60, pixel size = 2 mm × 2 mm, esp = 0.6 ms, L = 20, N = 16, and M = 36. A total of three trigger delays (18 ms, 28 ms, and 38 ms) were used, corresponding to a scan time of 108 heart beats (about 3.5 minutes including the idle time in between breath-holds). In the full FOV get-SPEEDI acquisition, the imaging parameters were identical to those used in the rFOV-SPEEDI sequence, except for TR = 20 ms, TE = 3 ms, FOV = 24 cm × 24 cm, reconstruction matrix size = 118 × 118, L = 32, and M = 80. In addition, two trigger delays (12 and 22 ms) were used in full FOV get-SPEEDI due to the shorter RF pulse width compared with the 2D RF pulse employed in rFOV-SPEEDI, resulting in a total scan time of 160 heart beats (about 6 minutes including the idle time in between breath-holds). Unlike the phantom experiment, the spatial resolution from the two sequences was kept identical to facilitate comparison.

2.3. Image Reconstruction and Analyses

In both experiments, the acquired k-space data from the full FOV get-SPEEDI and rFOV-SPEEDI sequences were reconstructed offline using customized MATLAB programs (MathWorks, Inc., Natick, MA). Images from each coil channel were first reconstructed, followed by a sum-of-square combination.

For the first experiment, phase difference image reconstruction was performed at each time point along the echo-train using a standard algorithm described in (18). From the phase difference maps, the phase evolution curves as a function of time were obtained pixel-by-pixel, followed by a first-order time derivative to obtain a quantity proportional to the applied current amplitude. To quantitively compare the measured signal S_acq and the applied signal S_app, a normalized root-mean-square error (NRMSE) was evaluated using:

$$\mathit{\text{NRMSE}}=\frac{\sqrt{\frac{1}{n}{\sum }_j{\left(S{\left(j\right)}_{\mathit{\text{acq}}}-S{\left(j\right)}_{\mathit{\text{app}}}\right)}^2}}{\mathbf{\text{max}}\left({\mathit{S}}_{\mathit{\text{app}}}\right)-\mathbf{\text{min}}\left({\mathit{S}}_{\mathit{\text{app}}}\right)},$$ [1]

where n is the number of samples (i.e., time points) in the phase map, S(j)_acq and S(j)_app are the jth sample of the experimentally measured (S_acq) and the applied (S_app) signals, respectively.

For the second experiment, the reconstructed images were reordered according to their time stamp in relation to the simultanously acquried ECG. To monitor the dynamic change of the aortic valve, the planimetric arotic valve area (AVA) was extracted from each image in the time series using a semi-automatic method described in (19). This was followed by generating the AVA curve corresponding to full FOV get-SPEEDI and rFOV-SPEEDI, respectively. The contrast-to-noise ratio (CNR) between the aorta and the right atrium and the signal-to-noise ratio (SNR) in the right atrium were evaluated for the two sequences to enable a quantitative comparison (see details in Supporting Information).

3. Results

3.1. First Experiment: Capturing the Dynamics of Fast-changing Currents

Figure 3 shows the time evolution of the phase difference maps with a full FOV covering the entire phantom (get-SPEEDI; Figure 3A) and with a reduced FOV focusing only on the central region of the phantom (rFOV-SPEEDI; Figure 3B). From both phase difference maps, the temporal phase evolution resulting from the action-potential-mimicking current pulse can be well visualized. Compared to the full FOV get-SPEEDI acquisition, the reduced FOV acquisition halved the scan time while offering a similar temporal resolution (0.8 ms vs. 0.6 ms) and a higher spatial resolution (pixel linear dimension = 1.25 mm vs. 1.88 mm).

Figure 3:

A set of phase difference maps acquired with full FOV using get-SPEEDI (A) and reduced FOV using rFOV-SPEEDI (B) in the loop wire experiment. The coverage of the reduced FOV is indicated by the red box in (A). The temporal resolution in full FOV get-SPEEDI and rFOV-SPEEDI was 0.6 ms and 0.8 ms, respectively. The number at the upper left corner of each image indicates the sequence in the time-series. For the action-potential-mimicking current waveform (black), the normalized experimental current evolutions at a randomly selected point near the dipole (red arrow) are shown in (C) for both full FOV get-SPEEDI (red) and rFOV-SPEEDI (blue). Each experimental waveform was averaged four times to reduce the adverse effects from noise. Both experimental current waveforms matched well with the input current signal (black), as evidenced by low NRMSEs (0.146 for full FOV get-SPEEDI and 0.148 for rFOV-SPEEDI). For the rectangular current waveform with a higher frequency (2 ms ON, 3 ms OFF; 200 Hz; blue), the corresponding experimental results are shown in (D). The phase evolution curves for full FOV get-SPEDI (red) and rFOV-SPEEDI (green) both matched well with the theoretical phase evolution (black dash line) calculated from the input current in blue (NRMSE = 0.444 for full FOV get-SPEEDI and NRMSE = 0.464 for rFOV-SPEEDI).

Figure 3C shows the current waveforms measured from the time evolution of the phase difference maps with a temporal resolution of 0.6 ms and 0.8 ms using full FOV get-SPEEDI and rFOV-SPEEDI, respectively. The experimentally obtained current waveforms matched well with the applied current signal (NRMSE = 0.146 and 0.148 for full FOV get-SPEEDI and rFOV-SPEEDI, respectively). Figure 3D displays the phase evolution curves obtained using the rectangular current waveform with a higher frequency of 200 Hz. Compared to the theoretical phase evaluation, full FOV get-SPEEDI and rFOV-SPEEDI yielded an NRMSE of 0.444 and 0.464, respectively.

3.2. Second Experiment: Visualizing Rapid Opening and Closing of the Aortic Valve

Figures 4A and 4B illustrate two sets of time-resolved images, both with a temporal resolution of 0.6 ms, during the rapid opening (Figure 4A) and rapid closing phases (Figure 4B) of the aortic valve from a healthy human subject. The dynamics was clearly visualized from both full FOV (top row) and reduced FOV acquisitions (bottom row). The CNR of aorta to right atrium was lower (~33) in the reduced FOV acquisition than in its full FOV counterpart (~42). The SNR in the right atrium exhibited a similar decrease (118.9 vs. 146.9). The CNR and SNR degradations, however, were counterbalanced by a scan time reduction of 32.5% (108 vs. 160 heart beats).

Figure 4:

Multiple frames in a time series showing the dynamics during the rapid opening (A) and rapid closing (B) of the aortic valve (indicated by the blue arrows) from a healthy human subject (male; 30 years of age) with full FOV acquisition (top) and reduced FOV acquisition (bottom) in each sub-figure. In both full and reduced FOV data sets, the temporal resolution was 0.6 ms. The full FOV images are magnified within the red box shown on the left. The same box was also used for prescribing the reduced FOV during the image acquisition.

Figure 5 displays two planimetric AVA dynamic curves obtained from the full FOV acquisition (red) and reduced FOV acquisition (blue) for the subject shown in Figure 4. The two AVA curves matched well with each other. From either AVA curve, the three phases during the aortic valve opening and closing can be well identified: a rapid opening phase (39 ms), a slow closing phase (279 ms), and a rapid closing phase (35 ms). In addition, both AVA curves revealed an overshoot when the aortic valve opened maximally, which was reported in the literature using non-MRI techniques (20,21).

Figure 5:

Changes in the planimetric aortic valve area (AVA) as a function of time, as measured by full FOV get-SPEEDI (red) and rFOV-SPEEDI (blue), both with a temporal resolution of 0.6 ms. The simultaneously acquired ECG waveform (light blue) was used as a trigger with the thin red vertical line (t = 0) indicating the trigger point. The two AVA curves matched well with each other. An overshoot (thick black arrow) was observed with both sequences when the aortic valve opened maximally. Five representative images acquired at different time points along the AVA curve are shown with the acquisition time of each image indicated in the parentheses (relative to the ECG-trigger).

4. Discussion

We have shown that reducing the FOV is a viable approach to shortening the scan times in SPEEDI. Using rFOV-SPEEDI, we have experimentally observed the fast-changing electric currents in a phantom and the dynamics of aortic valve rapid opening and closing in human subjects, both with sub-millisecond temporal resolutions in a total scan times of 1 to 3.5 minutes.

Despite their ability to provide ultrahigh temporal resolution, the previously proposed SPEEDI sequences (2,3) require relatively long scan times, which imposes a barrier for practical adoption. The relatively long scan duration in SPEEDI originates from its reliance on phase-encoding. This reliance is relaxed in get-SPEEDI where spatial encoding can be performed using a combination of frequency and phase encodings (2). Even with get-SPEEDI, the acquisition times can still be long for applications involving human subjects. For example, to image the human aortic valve opening and closing, get-SPEEDI required 8 to 10 breath holds (160 heart beats). rFOV-SPEEDI shortened the scan times by reducing the number of phase-encoding steps over a restricted FOV without compromising the spatial resolution. In the example of imaging the aortic valve dynamics, only ~5 breath holds (108 heart beats) were required to complete the image acquisition, representing a reduction of 32.5% in scan times. The reduction in scan times can be greater when the idle time between breath holds is also considered. Similarly, the duration of the experiment for measuring the dynamics of electric currents was decreased by two-fold without degrading the measurement accuracy. Additional reduction in the total scan times is possible by combining rFOV-SPEEDI with sparse k-space sampling methods such as compressed sensing (CS) (22), low-rank (23), sparse matrix decomposition (24), and machine learning (25).

In both get-SPEEDI and rFOV-SPEEDI, the temporal resolution was determined by esp. To adequately capture the dynamics of the aortic valve movement, we set esp to 0.6 ms to be consistent with the typical temporal resolution (i.e., 0.5 – 1 ms) used in M-mode echocardiography for aortic valve imaging. With this temporal resolution, fine features such as the overshoot (20,26) were successfully visualized. We recognize that the aortic valve opening and closing process may be too slow to take full advantage of the sub-millisecond capability of rFOV-SPEEDI. The results in the electric current experiments (Figure 3D), however, illustrate that a faster event (i.e., 200 Hz) can indeed benefit from the ultra-high temporal resolution afforded by rFOV-SPEEDI. The full benefits of sub-millisecond temporal resolution should be further investigated in faster periodic motions with a frequency of 1 kHz or higher.

Common to both full FOV get-SPEEDI and rFOV-SPEEDI, motion induced phase can propagate throughout the readout echo train for moving spins. Such phase errors have implications on both temporal and spatial resolutions. Using a nominal velocity of 60 cm/s for aortic valve movement, the motion induced phase accumulations at the first echo in the echo train were estimated to be ~53° and ~12° along the phase-encoding and frequency-encoding directions, respectively. These phase accumulations were likely responsible for the blurred edge of the aortic valve images, particularly during the rapid opening phase.

In rFOV-SPEEDI, we employed a 2D RF to limit the FOV. One potential issue with 2D RF pulse is the off-resonance signal from lipids, which leads to a spatial shift pre-dominantly along the blip-gradient direction. The lipid signals can be filtered out by using a 180° refocusing pulse applied to water signal at a specific location (27). Introducing a 180° refocusing pulse, however, can substantially increase the echo time. This can lead to missing the rapid opening phase of the aortic valve, which occurs about 30–50 ms after the QRS complex. For this reason, a 180° RF pulse was not incorporated into the rFOV-SPEEDI sequence in this study. Instead, the lipid signals detected predominantly by the few posterior coil elements were filtered out during image reconstruction. This simple approach worked well as evidenced by the good image quality shown in Figure 4.

One drawback of rFOV-SPEEDI is lower CNR and SNR as compared to full FOV get-SPEEDI. This was likely caused by a longer echo time in rFOV-SPEEDI as a result of an increased excitation RF pulse width (i.e., 14.8 ms), making the signal more sensitive to T2* decay. Another drawback is the sensitivity to eddy current perturbations during traversal of excitation k-space. Eddy currents with short time constants can distort the excitation k-space, and hence the spatial response of the 2D RF pulse (28). This issue was mitigated by using a “fly-back” gradient waveform (Figure 1B). It is worth noting that eddy currents with long time constants may also affect rFOV-SPEEDI because long-time-constant eddy currents induced by the fly-back gradient may propagate throughout the echo train and may even affect the following acquisitions. These effects require further investigation.

In conclusion, we have illustrated a time-efficient SPEEDI technique – rFOV-SPEEDI – that can provide high spatiotemporal resolution over a limited FOV. The experimental visualization of the fast-changing electric currents and the dynamics of human aortic valve represents two examples of using rFOV-SPEEDI for studying ultra-fast, cyclic biophysical and physiological processes with sub-millisecond temporal resolution within a practically acceptable time. With further development to address issues such as motion-induced phase variations, lipid suppression, CNR, SNR, and eddy currents, the rFOV SPEEDI technique is expected to help expand SPEEDI applications to visualizing the dynamics of a focal region within a larger object.