SHILO, a novel dual imaging approach for simultaneous HI-/LOw temporal (Low-/Hi-spatial) resolution imaging for vascular dynamic contrast enhanced cardiovascular magnetic resonance: numerical simulations and feasibility in the carotid arteries

The SHILO sequence is based on a segmented, 2D, T1-weighted radio frequency (RF) spoiled gradient recalled echo (SPGR), turbo field echo (TFE) acquisition. The SHILO segmentation/interleaving scheme is shown in Figure 1. To allow for different time resolution between AIF and tissue acquisitions, the AIF is acquired as a single shot low spatial/high temporal resolution image (Figure 1, L1-4) while the tissue is simultaneously acquired as a high spatial/low temporal resolution multi-shot image (Figure 1, H1A-D). Therefore complete AIF frames are interleaved with tissue image segments. Spreading the acquisition of the tissue image over n segments, allows the time resolution of the single shot AIF image to be (n + 1)/2 times faster than if the tissue image was also acquired as single-shot, as in previously proposed dual-imaging schemes [15, 37]. SHILO imaging parameters can be found in Table 1. Acquisition time for one segment is 0.8 s (TR × TFE factor). One AIF image is acquired every 1.6 sec, while one complete tissue image is acquired every 6.4 sec. AIF and tissue slices can be placed independently, and were typically separated by 15–20 mm in this case. A non-selective saturation pulse is played before every single-shot AIF acquisition (Figure 1, black bar). Similar to Kim et. al [15], to allow for different image contrast and dynamic signal range between AIF and tissue acquisition, a short saturation delay, τ = 0 ms, was used for AIF acquisition for increased sensitivity to the high concentration in the vessel lumen, while a longer delay, τ = 0.8 s, was used for tissue acquisition to allow for increased signal recovery before imaging. A repetition time (TR) of 10 ms was chosen to trade-off increased temporal resolution at lower TR with decreased signal saturation of longer T1 signal in the vessel wall at longer TR. Linear k-space reordering and a low flip angle of 15° was used for the AIF acquisition, similarly to Gatehouse et al. [37]. For the tissue acquisition, centric reordering was chosen so that contrast depends on recovery since the preceding saturation pulse rather than SPGR contrast. A higher flip angle of 25° was chosen to increase the signal excited at the center of k-space from the prepared magnetization.

Here we show that accuracy and precision of kinetic parameters related to atherosclerotic plaque microvascularization can be improved by sampling the AIF at faster time resolution than the plaque enhancement. This feature is implemented in the novel SHILO sequence by interleaving the low spatial/high temporal resolution AIF images with high spatial/low temporal resolution tissue acquisitions. A model AIF [43] with time resolution 0.1 s over a 5 minutes interval was used to compute tissue uptake curves using a modified Tofts-Kermode [44] model and a range of kinetic parameters, representative of vessel wall perfusion indices reported in the literature [45, 46]. Kinetic parameters used in the simulation were v _p , the fraction of intra-vascular volume; K ^trans , expressing the inflow of contrast agent from the plasma to the tissue compartment; v _e , the fraction of extra-vascular extra-cellular space. The parameter K _ep expressing the backflow of contrast agent from the tissue to the plasma compartment was calculated as K ^trans / _ve . 150 different combination of kinetic parameters were used, with v _p values 0.001, 0.005, 0.02, 0.05, 0.1, K ^trans values 0.02, 0.06, 0.1, 0.15, 0.2 min^-1, and v _e values 0.1, 0.2, 0.3, 0.4, 0.6, 0.8. Two different sampling schemes were then implemented while fitting for kinetic parameters [16]: 1) sampling of both AIF and tissue curves at the same increasingly lower rate (referred to as “same time resolution”, STR) to emulate deriving the AIF curve from the same tissue frames 2) sampling of the tissue curves only (referred to as “different time resolution”, DTR) at an increasingly lower rate to emulate the SHILO dual imaging or other dual bolus techniques. These different levels of under-sampling simulate situations in which an increasing number of tissue slices is acquired using both sampling schemes. In the STR scheme both AIF and tissue curves were under-sampled at rates 1.6, 3.2, 6.4, 12.8, 25.6 and 51.2 s (corresponding to multiples from 1 to 32 of the AIF time resolution in SHILO). In the DTR scheme, tissue curves were under-sampled at the same rate used for the STR case. As for AIF, two DTR schemes were investigated with the AIF sampled at 1.6 s and 0.8 s. The first case (1.6 s AIF sampling) corresponds to the temporal resolution of the AIF in the proposed SHILO acquisition. The second case correspond to a hypothetical test bolus acquisition, where the time resolution of the AIF is equal to the acquisition time of one single shot image as mentioned in the above description of the SHILO sequence, without interleaving the acquisition of segments of tissue slices. Comparing both these DTR schemes allows assessing the impact on kinetic parameters estimation of interleaving AIF and tissue acquisitions, such as is proposed in the SHILO sequence, as opposed to the corresponding test-bolus acquisition. For each parameter set and sampling scheme, the simulation was repeated 6 times, each time shifting the sampling grid by one sixth of the sampling interval to account for a variable temporal registration of the image sampling and bolus passage. Finally, kinetic parameters were estimated with the same modified Tofts-Kermode model used to simulate tissue curves [44], using non linear least squares fitting routines implemented in Matlab (MathWorks, Natick, MA). The average and standard deviation of estimated kinetic parameters across the 6 variable temporal registration cases were recorded. Relative errors, normalized by the true parameter value were recorded and the magnitude taken. The median of relative magnitude errors and standard deviation over the span of kinetic parameters simulated was plotted as a function of tissue slice temporal sampling resolution.

The equivalency of the MR signal of the newly developed SHILO sequence with its low and high spatial resolution standard SPGR counterparts was investigated using phantom experiments. Ten vials containing different dilutions of Gd-DTPA in saline with concentrations between 0 to 5 mM were imaged using the SHILO acquisition and using standard saturation-prepared 2D-SPGR acquisitions equivalent to the single-shot low-resolution and 4-fold segmented high-resolution SHILO images. For the high resolution SPGR image a saturation delay of 0.8 s was used as in the SHILO acquisition. SHILO imaging parameters were the same as described above (Table 1). An additional proton-density-weighted image (PD-w) was acquired with the same imaging parameters described in Table 1 except TR/TE 300/2.3 ms and flip angle = 4° to normalize the signal. Regions-of-interest were drawn in each vial using custom made Matlab software (MathWorks, Natick, MA). Normalized signal from SHILO and SPGR images were plotted against each other and compared using Pearson’s correlation coefficient.

In this experiment we verified that the AIF signal measured by the SHILO sequence was comparable to the AIF measured with an equivalent standard SPGR sequence in vivo. With approval from the Mount Sinai Institutional Review Board (IRB) and after written informed consent, four patients with either risk factors or history of atherosclerosis were recruited for carotid imaging with SHILO. Subjects were imaged on a 3T whole body scanner (Philips Achieva) using a dedicated 8-channel array for carotid imaging. After anatomical fast gradient echo and time-of-flight (TOF) angiographic scout images, standard T1, T2 and PD weighted turbo spin echo (TSE) black blood images were acquired for characterization of the vessel wall of both common carotid arteries using imaging parameters described in previous studies [47]. For normalization of the SHILO MR signal, an rf-spoiled 2D-SPGR image was acquired with TR/TE 300/2.3 ms and flip angle = 4° to give proton-density image weighting (PD-w) independent of T1 and T2 in each of the locations of the AIF and tissue uptake slices. AIF and tissue slices for SHILO DCE-CMR were then positioned in the common carotid arteries, perpendicular to the vessels. The tissue slice was positioned in the location with greatest vessel wall thickness as identified on multi-slice black blood TSE scans. The AIF slice was positioned 15–20 mm caudally with respect to the tissue slice. To confirm in vivo equivalency between the SHILO MR signal and the corresponding standard SPGR acquisition, subjects were injected twice, with a low dose of contrast agent, in the same imaging session. This approach allows minimizing the possible confounding effects of inter-scan variability. After placing an 18-gauge catheter in an antecubital vein, 0.01 mmol/kg of Gd-DTPA (Magnevist) was injected both times at a rate of 4 ml/s, followed by 20 ml saline flush (Figure 2). Contrast agent injection started after the acquisition of 3 high spatial resolution tissue images of the SHILO- DCE sequence (approximately 19.2 sec). During the first low dose injection, subjects were imaged using a low resolution SPGR sequence, equivalent to the AIF SHILO images. During the second low dose injection subjects were imaged using SHILO. Following acquisition, images were transferred to a workstation for image analysis. After image registration and de-noising [48] image signals were normalized by the PD-w images. One ROI was drawn in the vessel lumen of each common carotid for both the low resolution SHILO and SPGR AIF images. The average ROI signal intensity versus time was converted to T1 by using the SPGR signal equation, and converted to concentration of gadolinium in the blood [Gd]_blood using [Eq. 1]. Contrast agent relaxivity was assumed to be r = 4.2 s^-1mM^-1. The initial value of T_1,0 was taken from the average of pre-contrast images before the influx of contrast agent. For kinetic analysis, the concentration [Gd]_blood was corrected for hematocrit.

Correspondence between AIF curves derived from SHILO and standard SPGR acquisitions was evaluated by visual inspection. AIF curves from all vessels of each subject were averaged after appropriate translation in time to overlay the initial bolus arrival. The Pearson’s correlation coefficient for this set of data points was found.

In the same cohort of patients, and in the same imaging session, after two low dose injections for in vivo validation of SHILO, an additional higher dose injection was used to assess feasibility of using SHILO for perfusion studies in carotid atherosclerosis (Figure 2). For this purpose, 0.05 mmol/kg of Gd-DTPA (Magnevist) was injected at a rate of 4 ml/s, followed by 20 ml saline flush. In this case also, contrast agent injection started after the acquisition of 3 high spatial resolution tissue images of the SHILO- DCE sequence (approximately 19.2 sec). Following acquisition, images were transferred to a workstation for image analysis. After image registration and de-noising [48] image signals were normalized by the PD-w images. One ROI was drawn in the vessel lumen of each common carotid for the low resolution AIF image, while for the high resolution tissue image ROIs encompassed the whole visible vessel wall of both common carotids and areas in the sternocleidomastoid muscle. The average ROI signal intensity versus time for both AIF and tissues was converted to T1 by using the SPGR signal equation, and converted to concentration using [Eq. 1]. Contrast agent relaxivity was assumed to be 4.2 s^-1mM^-1. The initial value of T_1,0 was taken from the average of pre-contrast images before the influx of contrast agent. AIF concentration curves were corrected for hematocrit before kinetic modeling. A modified Tofts model [44] was used to analyze concentration curves, by using the AIF and tissue data from SHILO images. Kinetic parameters v_p (fraction of vascular space), K^trans (wash-in constant from plasma to tissue compartment), v_e (fraction of extravascular extracellular space), and K_ep (K^trans/_ve wash-out constant from tissue to plasma compartment) were calculated using standard non-linear least squares fitting procedures [49] written in Matlab.

Figure 3 shows the effect of under-sampling at different rates for the AIF curve (Figure 3, panel A) and three different tissue curves (Figure 3, panels B-D). Red solid lines represent the fully sampled data, while green and blue lines represent under-sampled data (8 and 32 times respectively). The plot demonstrates the dramatic effect of under-sampling on the amplitude of the AIF peak, as well as on the initial portion of fast enhancing curves (Figure 3, panels C and D). The effect of under-sampling on slower enhancing curves (Figure 3, panel B) is less evident. Results of the simulations are shown in Figures 4 and 5 for estimations of each of the kinetic parameters separately. Figure 4 shows the median of relative magnitude errors, while Figure 5 shows the median standard deviation over the span of kinetic parameters, plotted as a function of tissue curves sampling resolution. Errors and standard deviations for both DTR and STR schemes increase as the tissue temporal resolution is decreased, thus confirming the results of previous studies [16, 17]. The log scale for relative error and standard deviation shows that DTR schemes vastly outperform the STR scheme, especially at lower temporal resolution (Figures 4 and 5). Also evident is the close similarity of the two DTR schemes with AIF temporal resolutions of 0.8 sec and 1.6 sec. This indicates that the penalty paid when interleaving AIF and tissue images in the SHILO sequence does not substantially affect the estimation of kinetic parameters with respect to the corresponding dual bolus acquisition. The accuracy and precision of kinetic parameter estimation with the DTR schemes was better than 1%. The median error, over the span of kinetic parameter sets simulated, for all tissue sampling rates was less than 1% (Figure 4). Similarly, the median standard deviation of the kinetic parameter estimation was less than 1% (Figure 5). Conversely, accuracy and precision of parameter estimation with the STR scheme was typically between 1% and 100%, as shown by the median errors and standard deviations shown in Figures 4 and 5.

There is a close correspondence between SHILO and standard techniques for each of the comparisons: i) low-resolution SHILO (SHILO-LO) versus single-shot SPGR, and ii) high resolution SHILO (SHILO-HI) versus segmented SPGR with a 0.8 s saturation delay (Figure 6). Measured signal from SHILO is plotted against those from SPGR in Figure 6 and shows a correlation with R² = 0.99. This indicates that the modifications required for interleaving AIF and tissue scans do not compromise the fidelity of signal acquisition required for accurate T1 measurements in the SHILO DCE scan.

There is a striking correspondence between the AIF shapes acquired with the SHILO sequence and the standard SPGR equivalent. Shown in Figure 7 are the AIF curves from SHILO and SPGR plotted on the same axes for (A) the average of all AIFs from all vessels in all subjects, and (B) for one single vessel. The correlation coefficient between the averages AIFs for SHILO and SPGR was R² = 0.95 (slope 0.95, intercept - 0.01).

For a representative subject, AIF and tissue images taken from the SHILO sequence are shown in Figure 8A and B and indicate the image quality obtainable. Concentration-time curves for the AIF, vessel wall, and muscle measured from ROIs placed on the dynamic series of these images are plotted in Figure 8C. The AIF curve clearly shows the passage of the contrast agent bolus [43], and the tissue curves are consistent with shapes expected for low perfusion tissue [5, 6, 19, 50‐52]. Average kinetic parameter values for the vessel wall were respectively: K_trans : 0.27 ± 0.26 min^-1; v_p: : 0.02 ± 0.03; v_e: 0.53 ± 0.26 . Average kinetic parameters values for the skeletal muscle were respectively: K_trans : 0.09 ± 0.04; min^-1 v_p: : 0.00 ± 0.00; v_e: 0.27 ± 0.06. These values compare favorably with similar estimates found in the literature [6, 51‐53].

This feasibility study presents several limitations. As shown by numerical simulations, some advantages of the DTR sampling scheme implemented in the SHILO sequence can be mostly appreciated when the disparity between the AIF and tissue time resolution is significant. This is the case when imaging more than one tissue slice, simultaneously with one AIF slice. In this feasibility study we omitted extensive exploration of comparison with acquisitions of multiple tissue slices, instead demonstrating the technique with a single tissue slice. Our aim was to demonstrate the feasibility of applying the DTR sampling scheme, while providing physical separation between AIF and tissue slices and independence of imaging parameters, attributes not available in other k-space sharing techniques. Comparison of tissue curves between SHILO and SPGR in the low dose scans was not possible, because of the too low uptake of contrast agent in the vessel wall. Addition of a second high-dose to make such a comparison would have increased total patient dose beyond the recommended guidelines. Finally, the small number of patients included and the lack of histological samples limits the extent of the assessment of the SHILO technique. However, we feel that the data presented in this feasibility study will warrant the future investigation of this novel approach in a more extensive study, where the absolute accuracy of plaque perfusion parameters will be compared against histological features as gold standard.