Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2012 Jul 10.
Published in final edited form as: Magn Reson Med. 2011 Jan 10;65(5):1269–1277. doi: 10.1002/mrm.22717

3D Radial Sampling and 3D Affine Transform-based Respiratory Motion Correction Technique for Free-breathing Whole-Heart Coronary MRA with 100% Imaging Efficiency

Himanshu Bhat 1, Lan Ge 1, Sonia Nielles-Vallespin 2, Sven Zuehlsdorff 3, Debiao Li 1,4,*
PMCID: PMC3202653  NIHMSID: NIHMS310811  PMID: 21500255

Abstract

The navigator gating and slice tracking approach currently used for respiratory motion compensation during free-breathing coronary magnetic resonance angiography (MRA) has low imaging efficiency (typically 30–50%), resulting in long imaging times. In this work, a novel respiratory motion correction technique with 100% scan efficiency was developed for free-breathing whole-heart coronary MRA. The navigator signal was used as a reference respiratory signal to segment the data into six bins. 3D projection reconstruction k-space sampling was used for data acquisition and enabled reconstruction of low resolution images within each respiratory bin. The motion between bins was estimated by image registration with a 3D affine transform. The data from the different respiratory bins was retrospectively combined after motion correction to produce the final image. The proposed method was compared with a traditional navigator gating approach in nine healthy subjects. The proposed technique acquired whole-heart coronary MRA with 1.0 mm3 isotropic spatial resolution in a scan time of 6.8 ± 0.9 min, compared with 16.2 ± 2.8 min for the navigator gating approach. The image quality scores, and length, diameter and sharpness of the right coronary artery (RCA), left anterior descending coronary artery (LAD), and left circumflex coronary artery (LCX) were similar for both approaches (P > 0.05 for all), but the proposed technique reduced scan time by a factor of 2.5.

Keywords: coronary MRA, 3D PR k-space sampling, 3D affine transform, respiratory motion correction

Whole-heart coronary MRA acquired during free-breathing (1,2) is a promising technique for noninvasively detecting coronary artery disease (3,4). Currently, the most common method of dealing with respiratory motion during free-breathing is to acquire a diaphragmatic navigator and use a fixed correlation factor (usually 0.6) to estimate and correct for the translational motion of the heart in the superior–inferior direction (5,6). All other directions (i.e., anterior–posterior and left–right) and types (rotation, scaling, etc.) of motion are neglected. This approach works well because of the small gating window of 5–6 mm, which is used during the acquisition. However, this small window leads to low-imaging efficiency (30–50%) resulting in long image acquisition times.

Several methods have been proposed to increase the acceptance window size and, thereby, the scan efficiency. Almost all of them involve using some sort of patient-specific information in the motion model, which allows for more accurate motion detection and correction over a wider range of heart motion. One of these methods uses a patient-specific correlation factor and corrects for the superior–inferior translational motion of the heart (7). Another class of methods attempts to improve the navigator gating algorithm over the simple accept/reject approach (8,9). Other methods use low-resolution calibration scans to fit more complicated motion models to each individual patient (10). A 3D translational model, which models the translation in the superior–inferior, left–right, and anterior–posterior directions, and a 3D affine transform model, which models the rotation and scaling of the heart in addition to the three translations, have been used with this technique. Self-gating, which directly measures and corrects for translational heart motion (either 1D of 3D) by acquiring projections through the heart has also been used with both Cartesian and radial k-space data sampling (11,12). Fat-navigator methods, which acquire a full 3D image of the fat surrounding the heart and use this information for beat-to-beat correction of heart position (13–15) have also been proposed.

In this article, we propose a new method for respiratory motion correction in whole-heart coronary MRA with 100% imaging efficiency. The traditional navigator signal (on the right hemidiaphragm) is acquired during data acquisition but is not used for any gating or motion correction. It is instead used as a reference respiratory signal to segment the data into several respiratory bins, where each bin represents a particular state in the respiratory cycle. A 3D projection reconstruction k-space trajectory is used for the data acquisition. This enables reconstruction of low resolution images with under-sampled data in each respiratory bin. Image registration is used to estimate the motion between the different bins. The motion model used is a 3D affine transform, which accounts for rotation, scaling, and translation of the heart. Data from all the respiratory bins is retrospectively combined after motion correction and the final high-resolution image is obtained. Healthy volunteer studies and numerical simulations were performed to evaluate the effectiveness of this technique. The images reconstructed with the proposed technique were compared with the traditional navigator gating and slice tracking approach (5,6).

MATERIALS AND METHODS

Details about the motion estimation, image reconstruction with motion correction and 3D projection reconstruction (PR) k-space sampling are given below.

Motion Estimation

A schematic of the motion estimation procedure is shown in Fig. 1. The navigator signal is first segmented into six evenly spaced bins. As shown in Fig. 1, bin 1 is the end expiratory bin, bin 6 is the end inspiratory bin and bins 2–5 are in between. The k-space data is then split according to respiratory bins. The k-space sampling pattern in each respiratory bin cannot be predicted and is both nonuniform and undersampled. To minimize artifacts due to these effects, the k-space data for each bin is low-pass filtered using a 3D Gaussian kernel. This is followed by gridding (16) to get the low-resolution 3D images for each respiratory bin. A spatial mask is then applied to restrict the registration in the region of the heart. The next stage in the motion estimation is the iterative 3D image registration. The registration framework used in this work was based on the Insight Tool Kit (www.itk.org) (17) which is an open-source toolkit. The registration framework (shown in the dotted box in Fig. 1) consists of a reference image and a target image (which is iteratively registered to the reference image). In this work, bin 1 (the end expiratory bin) was used as the reference image, and bins 2–6 were sequentially used as the target image. The transform is the spatial mapping from the target image space into the reference image space. In this work, the 3D affine transform was used to estimate the rotation, scaling, shearing, and translation of the heart between the different respiratory bins. The 3D affine transform is represented as:

[x′y′z′1]=[M00M01M02TxM10M11M12TyM20M21M22Tz0001]·[xyz1] [1]

where [x′, y′, z′] is the transformed position vector, [x y z] is the original position vector, [Tx, Ty, Tz] is the 3D translation vector, and the matrix represents the linear transformation, which combines the rotation, scaling and shearing components of the affine transform. Linear interpolation was used to estimate the transformed target image at nongrid positions. The metric used in the registration is the function which is minimized during the iterative process. In this work, we evaluated both the cross-correlation and mean squared error as metrics. The gradient descent algorithm was used for the optimization. During the registration the stopping criterion was either 200 iterations, or a minimum step size of 0.002, whichever was reached first. Using this registration framework, respiratory bins 2 to 6 were sequentially registered to bin 1, and the registration results were stored for further use during the motion correction procedure (described next).

FIG. 1.

FIG. 1

Open in a new tab

Schematic of the motion estimation procedure. The 3D PR data is segmented into six respiratory bins based on the navigator signal, where each bin represents a particular state in the respiratory cycle. 3D PR sampling enables reconstruction of low-resolution under-sampled images for each respiratory bin. The motion between the bins is estimated by image registration with a 3D affine transform, which estimates rotation, scaling, shearing, and translation between bins.

Image Reconstruction with Motion Correction

A schematic of the motion correction procedure is shown in Fig. 2. The k-space data is once again split into the same six respiratory bins used during motion estimation. All the bins undergo gridding to get the 3D images for each bin. The affine transform (estimated for each bin during Motion estimation) is then applied to the complex data in each bin. During motion correction, windowed sinc interpolation, which is both more accurate and computationally intensive compared with linear interpolation, is used to estimate the transformed image at nongrid positions. The hamming window is used to truncate the sinc kernel. The complex image data from all the bins is then summed together to get the final motion corrected 3D image for a particular coil. This procedure is applied for each coil independently and sum-of-squares is used to combine data from multiple coils.

FIG. 2.

FIG. 2

Open in a new tab

Schematic of the motion correction procedure. The data from different respiratory bins is retrospectively combined after motion correction based on the affine transform obtained during motion estimation, to produce the high-resolution motion corrected image.

3D PR Trajectory

The 3D PR trajectory used in this work was based on that proposed in Ref. 18 and samples data on a spiral path running on the surface of a sphere. Following are the expressions for the normalized Gx, Gy and Gz gradients:

Gz(n)=n−0.5NGx(n)=cos(2NπMsin−1(Gz(n))+2mπM)1−Gz(n)2Gy(n)=sin(2NπMsin−1(Gz(n))+2mπM)1−Gz(n)2 [2]

where n = 1, 2…N, m = 1, 2…M, M is the number of spirals along the surface of the sphere, and N is the number of projections in each spiral. Based on these equations, multiple trajectories can be designed by varying M and N such that the product MN is always equal to the total number of projections. For M = 1, the sampling is uniform over the surface of the sphere, and as we increase M (keeping MN = constant) holes are created near the poles of the sphere leading to nonuniform sampling (18). The trajectory needed in this work must satisfy two conditions: (i) it must sample data over the entire sphere in each heartbeat: this is critical, since it gives the ability to reconstruct low-resolution images for each bin even if there is a drift in breathing pattern during imaging, and (ii) the gradients must not change drastically from repetition time (TR) to TR: this is to avoid eddy current related artifacts in steady state free precession (SSFP) imaging (19). Keeping these two factors in mind three different k-space trajectories were simulated, and the results are shown in Fig. 3. For this simulation, we used typical parameters of: total projections = 16,000 and segments (projections acquired in each heartbeat) = 40. In Fig. 3, the first column shows the data sampling pattern in a sample heartbeat, the second column shows the Gx (blue), Gy (red), and Gz (pink) gradients required in each heartbeat, the third column shows the difference between the gradients in successive TR’s (once again Gx is blue, Gy is red, and Gz is pink), and the last column shows a phantom image collected with that specific trajectory. The first trajectory uses M = 400 (total projections/segments) and N = 40 (segments) (2). The advantage of this trajectory is that the gradients change very little in successive TR’s, and so the eddy current related artifacts are minimized. However, the sampling in each heartbeat is limited to a very narrow region of the sphere and, hence, does not satisfy the first condition listed above. The second trajectory uses M = 1 and N = 16,000, with data acquisition in every heartbeat interleaved along the single spiral path. The advantage of this trajectory is that it has uniform sampling all over the surface of the sphere (18). However, it results in significant Gx and Gy gradient fluctuations in successive TR’s, resulting in artifacts in the phantom image (19). The third trajectory uses M = 16 and N = 1000, with data acquisition in every heartbeat interleaved along each of the 16 spiral paths on the sphere. The phantom images with this trajectory are similar to the M = 400 trajectory indicating that the slightly increased gradient differentials do not lead to increased eddy current artifacts in the images. The major advantage of this trajectory is that it samples data over the entire sphere in each heartbeat. As a result, this 3D PR k-space trajectory with M = 16 and N = 1000 was used in this work.

FIG. 3.

FIG. 3

Open in a new tab

Comparison between three 3D PR k-space trajectories. The first column shows the data sampling pattern in a sample heartbeat, the second column shows the Gx (blue), Gy (red), and Gz (pink) gradients required in each heartbeat, the third column shows the difference between the gradients in successive TR’s (once again Gx is blue, Gy is red, and Gz is pink), and the last column shows a phantom image collected with that specific trajectory. The third trajectory (M = 16 and N = 1000) was used in this work as it samples data over the entire sphere in each heartbeat, and the imaging gradients do not change drastically from TR to TR. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Evaluation of the Motion Correction Technique

Simulations

One of the healthy volunteer datasets acquired with the navigator gating and slice tracking approach (described in next section) was used in the simulations. To reconstruct the reference image, the k-space data acquired in 50 evenly spaced heartbeats across the entire acquisition was reconstructed as shown in Fig. 1. For the target image, the k-space data for the desired heartbeats was accumulated and processed as shown in Fig. 1. In addition, linear phase terms were applied in all three directions to simulate translational motion (the simulation was restricted to translational motion between the reference and target images). The superior–inferior translation ranged from 2 to 10 mm, the anterior–posterior from 2 to 4 mm and the left–right from 2 to 4 mm. Each target image with the known translation was then registered to the reference image as shown in Fig. 1. The result of the registration was compared to the known translation in terms of Euclidian distance, and the result was averaged across all the different combinations of known translations.

The goal of the simulation was to find the minimum number of heartbeats required in each respiratory bin to accurately estimate the motion between bins. There are two important parameters to consider. One is the number of heartbeats which contribute data to the respiratory bin, and the second is the distribution of these heartbeats in the acquisition. The best case situation will be if the contributing heartbeats are evenly spaced across the entire acquisition, as this will ensure that the projections are well spaced out on the surface of the sphere. The worst case situation will be if all the contributing heartbeats are consecutive, as this will mean that the data are not well distributed on the surface of the sphere. To account for this effect, both these conditions were simulated along with another condition in which the contributing heartbeats were randomly distributed across the acquisition. The number of heartbeats was varied and, for each case, the three different heartbeat distributions described above were simulated. The results were compared in terms of the Euclidean distance between the true and estimated translations.

Healthy Volunteer Studies

Nine healthy volunteers (seven men and two women, average age 26.4 ±4.0 years) were scanned on a clinical 1.5T scanner (MAGNETOM Avanto, Siemens AG Healthcare Sector, Erlangen, Germany). A 12 channel matrix coil (six anterior and six posterior) in triple mode was used for data acquisition, resulting in four channels in the raw data. Written consent was obtained from the volunteers in compliance with the Institutional Review Board of our institution. A T2-prepared SSFP sequence with 3D PR sampling was used to acquire whole-heart coronary MRA. Parameters for the sequence were: TR = 3.2 ms, echo time (TE) = 1.6 ms, 250 ±sec hard pulse with flip angle = 90°, readout bandwidth = 781 Hz/pixel, 15 preparation pulses in each heartbeat with linear flip angle modulation (20), T2-prep duration = 40 ms, chemically selective fat-saturation pulse, 25–50 lines per heartbeat in a data acquisition window of 80–160 ms, total projections = 16,000 to 16,800 representing an under-sampling factor ~ 6.4 compared with Nyquist (2,12), field of view (FOV) = 260 mm3, matrix size = 2563, and voxel size = 1.0 × 1.0 × 1.0 mm3 interpolated to 0.5 × 0.5 × 0.5 mm3. For all the scans, a four chamber cine scan was used to determine the quiescent period for coronary artery imaging. Navigator pulses were placed on the dome of the right hemidiaphragm. For the proposed motion correction technique, the navigator was acquired but not used for any gating or motion correction during imaging. This ungated scan was reconstructed using the proposed motion estimation and correction technique. For comparison purposes, it was also reconstructed without any motion correction. For comparison purposes, the same sequence was also run with the traditional navigator gating and slice tracking approach with a fixed correlation factor of 0.6 and a small acceptance window of 6 mm (5,6). For a fair comparison, the order of acquisition of the two sequences was randomized. The three techniques: (i) no respiratory gating with proposed motion correction, (ii) no respiratory gating without motion correction, and (iii) navigator gating and slice tracking with ±3 mm window, were compared in terms of visualized LAD, RCA, and LCX lengths, diameters and sharpness and overall image quality scores. Image quality was evaluated by two independent, experienced observers who were blinded to the technique. All image quality assessments were based on the raw images. The image quality was graded as 1, poor (nonassessable); 2, fair (mild to moderate artifacts); 3, good (minimum to mild artifacts); 4, excellent (minimum or no artifacts). To measure vessel length, each artery was manually traced over multiple slices and displayed in a single 2D image from which the vessel lengths were measured. Diameter and sharpness measurements were based on Ref. 21. ImageJ (Image Processing and Analysis in Java http://rsbweb.-nih.gov/ij/) was used for all the measurements. A two-sided paired t test was used for all the statistical analysis with a P value of 0.05 considered to be statistically significant. The CoronaVIZ software (22) (Siemens Corporate Research, Inc., Princeton, NJ) was used for projecting multiple vessel branches from the whole-heart coronary MRA datasets onto a single image for visualization purposes.

RESULTS

Simulations

Figure 4 shows the results of the second simulation. When the target image consists of data from a minimum of 40 heartbeats the error is always less than 1 mm independent of the distribution of these 40 heartbeats during data acquisition.

FIG. 4.

FIG. 4

Open in a new tab

Simulation result showing that when the target image consists of data from a minimum of 40 heartbeats the error is always less than 1 mm independent of the distribution of these 40 heartbeats during data acquisition.

Healthy Volunteer Studies

Quantitative comparison between the three techniques: (i) proposed motion correction method with no navigator gating, (ii) no motion correction with no navigator gating, and (iii) navigator gating and slice tracking, is shown in Table 1. The average scan time with the proposed motion correction technique was 6.8 ±0.9 min with 100% scan efficiency. The average scan time for the navigator gating and slice tracking approach was 16.2 ±2.8 min with 43.4 ±11.5% scan efficiency. The image quality scores and length, diameter and sharpness of the RCA, LAD, and LCX were similar for both the proposed motion correction and navigator gating and slice tracking approaches (P value > 0.05 for all). The image quality and length and sharpness of the RCA, LAD, and LCX were significantly lower (P value < 0.05) without any motion correction compared with both the proposed motion correction and navigator gating and slice tracking approaches.

Table 1.

Quantitative comparison between the three techniques.

Parameter No respiratory gating with proposed motion correction (MC) No respiratory gating with no motion correction (noMC) Navigator gating and slice tracking with ±3mm acceptance window (NGS) P value (n = 9)
MC versus NGS NGS versus noMC noMC versus MC
Imaging time (min) 6.8 ±0.9 6.8 ±0.9 16.2 ±2.8 <0.001 <0.001 –
Navigator efficiency (%) 100.0 ±0.0 100.0 ±0.0 43.44 ±11.5 – – –
Image quality score 3.25 ±0.32 1.87 ±0.46 3.36 ±0.40 0.084 <0.001 <0.001
RCA sharpness (mm−1) 0.84 ±0.07 0.68 ±0.13 0.81 ±0.10 0.238 0.003 0.002
RCA diameter (mm) 3.31 ±0.46 3.61 ±0.60 3.36 ±0.39 0.703 0.157 0.201
RCA length (cm) 11.63 ±2.18 9.08 ±2.84 12.11 ±1.82 0.256 0.004 <0.001
LAD sharpness (mm−1) 0.83 ±0.09 0.61 ±0.07 0.84 ±0.12 0.677 0.001 0.001
LAD diameter (mm) 3.23 ±0.74 3.54 ±0.95 3.28 ±0.76 0.782 0.355 0.344
LAD length (cm) 9.16 ±1.17 5.46 ±2.57 9.13 ±1.12 0.908 0.006 0.003
LCX sharpness (mm−1) 0.92 ±0.06 0.65 ±0.14 0.90 ±0.08 0.667 0.031 0.009
LCX diameter (mm) 2.80 ±0.49 3.28 ±0.19 2.90 ±0.44 0.338 0.051 0.050
LCX length (cm) 6.32 ±1.31 2.07 ±1.57 6.13 ±1.10 0.372 0.001 0.002
Open in a new tab

Values are reported as mean ±standard deviation, and all statistically significant P values are in bold.

Figure 5 shows the entire motion correction procedure in a healthy volunteer. The same reformatted slice (containing a portion of the LCX) is plotted across all the bins to demonstrate the efficacy of motion correction. Included are: (i) images of individual bins before motion correction, (ii) affine transform parameters applied to each bin, (iii) individual bin images after motion correction, and (iv) images of all the bins combined together with and without motion correction. The improvement in image quality after motion correction is evident.

FIG. 5.

FIG. 5

Open in a new tab

The entire motion correction procedure in a healthy volunteer. The same reformatted slice (containing a portion of the LCX) is plotted across all the bins to demonstrate the efficacy of motion correction. The improvement in image quality after motion correction is evident.

Figure 6 shows reformatted coronary artery images from five healthy volunteers with the three techniques. Without any motion correction (right column), the images are blurry and the coronary artery visualization is poor. With the proposed motion correction technique (middle column), coronary artery visualization is excellent and similar to the navigator gating and slice tracking approach (left column). The imaging time with the motion correction method is reduced approximately by a factor of 2.5 to 3 compared with the navigator gating and slice tracking approach.

FIG. 6.

FIG. 6

Open in a new tab

Reformatted coronary artery images from five healthy volunteers. With the proposed motion correction technique (middle column), coronary artery visualization is excellent and similar to the navigator gating and slice tracking approach (left column). Without any motion correction (right column), the images are blurry and the coronary artery visualization is poor. The imaging time with the motion correction technique is reduced by a factor of 2.5 to 3 compared with the navigator gating and slice tracking approach.

DISCUSSION

Summary of Proposed Motion Correction Technique

In this work a novel motion correction technique is developed for respiratory motion compensation during free-breathing whole-heart coronary MRA. The navigator signal is used as a reference respiratory signal to segment the data into six respiratory bins, where each bin represents a particular state in the respiratory cycle. 3D PR sampling is used for the data acquisition and enables reconstruction of low-resolution under-sampled images for each respiratory bin. The motion between bins is estimated by image registration with a 3D affine transform, which estimates rotation, scaling, shearing, and translation between bins. The data from the different respiratory bins is combined after motion correction to produce the high-resolution motion corrected image. A major advantage of the proposed method is its 100% scan efficiency, which results in significant reduction in scan time. Using the proposed technique, whole-heart coronary MRA with 1.0-mm3 isotropic spatial resolution was acquired in a scan time of 7 min.

In the current work, six respiratory bins were used since this resulted in a reasonable balance between the amount of data in each bin, reconstruction time, and the motion present within each bin. With six bins, the superior–inferior displacement in each bin (in terms of diaphragm motion) ranged from 2.17 to 4.5 mm. This is less than the 5- or 6-mm navigator window, which is typically used for coronary MRA, and should keep the motion within each bin in acceptable limits. The effects of within bin motion are also minimized by the inherent insensitivity of PR sampling to motion (23). The current study used a relatively long data acquisition window of 80–160 msec in each heartbeat depending on the heart rate. This will have to be carefully adjusted in patients with higher and more variable heart rates. However, once again the inherent insensitivity of the 3D PR sampling to motion could make it feasible to use relatively longer acquisition windows with this technique compared with traditional Cartesian acquisitions.

Comparison with Other Techniques

The idea of using low-resolution PR images for motion correction was originally proposed for 2D radial imaging (24). Image registration for respiratory/cardiac motion correction in the heart has previously been used with real-time imaging (25) and 2D radial imaging (26). In this work, we extended these ideas to 3D PR k-space sampling, and combined it with the navigator information to get image registration-based 3D affine correction of respiratory motion in the heart during free-breathing whole-heart coronary MRA.

Over the years, several techniques have been proposed for improved respiratory motion correction in coronary MRA. One of these methods use a patient-specific correlation factor based on calibration scans and corrects for the superior–inferior translational motion of the heart (7). Another method uses a more complicated calibration scan for patient-specific, prospective, 3D affine correction of respiratory motion (10,27). Both these methods have the advantage that the motion correction is applied prospectively. The downside of this is that the calibration scan and processing needs to be done during the scan itself, adding more complexity to the already involved coronary MRA scanning protocol. Also, the motion model, which is fit during the calibration scan, could change during the actual scan. In comparison, the proposed method does not change the scanning procedure at all and all the complexity in the technique is during the motion estimation and correction steps, which are retrospectively applied to the data.

Self-gating is another promising method for respiratory motion correction. It directly measures and corrects for translational heart motion (either 1D or 3D) by acquiring projections through the heart (11,12). The advantage of this method is that it is not computationally intensive and can potentially be applied prospectively. One of the issues with the self-gating method is that the projection through the body, in addition to the heart, contains stationary tissues like the chest wall, complicating the motion estimation. In comparison, the proposed motion correction technique directly uses the image for the motion estimation. As a result, the registration is constrained to the region of the heart and is not influenced by signals coming from stationary tissues. The self-gating method has also so far been restricted to translational motion correction.

Fat-navigator methods, which acquire a full 3D image of the fat surrounding the heart and use this information for beat-to-beat correction of heart position (13–15), have also been proposed. The major advantage of these methods is that a beat-to-beat correction of respiratory motion is achieved. This is currently not possible with the proposed method.

Limitations of the Technique and Future Work

One of the disadvantages of this technique is that it is computationally intensive. In the current implementation, the reconstruction is done in Matlab (Mathworks, Natick, MA), with C and C++ subroutines being called for the gridding and registration, respectively. On a PC with an Intel Xeon 3.0 GHz processor, for four channel data, the 3D registration takes approximately 90 minutes, and the image reconstruction takes approximately 25 minutes. Significant reduction in this time should be possible by more careful optimization of the programs and the use of parallel computing; however, this long processing time and large memory requirement are serious impediments in applying the current technique on data collected with high-density array coils. The heavy computational load also makes it difficult to steer the data acquisition based on real-time estimation of motion, limiting the technique to retrospective motion correction. Another disadvantage is that the spatial mask, which is used to limit the registration to the region of the heart is currently manually selected. This mask is important so that stationary tissues like the chest wall and bright areas like the stomach do not influence the registration. Another disadvantage of this technique is the lower signal to noise ratio (SNR) in radial sampling compared with Cartesian sampling (28).

Since the breathing pattern is different for each subject, the sampling pattern is different in each bin and for each scan. We currently use a fixed quadratic function for sampling density compensation for all the bins, independent of the sampling pattern. An iterative density compensation scheme (29) may be better suited for such a case and should be explored in the future. This study was limited to healthy volunteers and excellent image quality was achieved with the proposed motion correction technique with 100% scan efficiency. In the future, the efficacy of this technique needs to be tested in a patient population, who are bound to have more variable and drifting breathing patterns. To account for this, the proposed technique may have to be combined with some sort of prospective gating to reject outliers during the acquisition. Also, instead of accepting all the data during the scan, the proposed method may be combined with a smaller navigator gating window for more accurate motion correction. An interesting possibility is that of beat-to-beat correction of respiratory motion similar to that achieved in the fat navigator approach (15). Currently, we acquire 25–50 segments per heartbeat, and this is not enough for accurate motion determination on a beat-to-beat level. However, this should be explored in the future.

CONCLUSIONS

In conclusion, a novel respiratory motion correction technique was developed for whole-heart coronary MRA. This technique uses a 3D PR trajectory to sample the data, and a 3D affine transform to correct for respiratory motion of the heart. Using the proposed technique, whole-heart coronary MRA was acquired with 1.0-mm3 isotropic spatial resolution, in a scan time of approximately 7 min with 100% scan efficiency. In healthy volunteers, excellent coronary artery visualization was achieved. Clinical utility of the proposed technique needs to be tested on a patient population.

Acknowledgments

The authors thank Xiaoming Bi, PhD, for scoring the images. Himanshu Bhat acknowledges the support from the Captain and Mrs. Roberts Fellowship.

Grant sponsors: National Institute of Health grants nos. NIBIB EB002623 and NHLBI HL38698; Siemens Medical Solutions USA Inc. Malvern, PA; The Captain and Mrs. Roberts Fellowship.

References

  • 1.Weber OM, Martin AJ, Higgins CB. Whole-heart steady-state free precession coronary artery magnetic resonance angiography. Magn Reson Med. 2003;50:1223–1228. doi: 10.1002/mrm.10653. [DOI] [PubMed] [Google Scholar]
  • 2.Stehning C, Bornert P, Nehrke K, Eggers H, Dossel O. Fast isotropic volumetric coronary MR angiography using free-breathing 3D radial balanced FFE acquisition. Magn Reson Med. 2004;52:197–203. doi: 10.1002/mrm.20128. [DOI] [PubMed] [Google Scholar]
  • 3.Sakuma H, Ichikawa Y, Chino S, Hirano T, Makino K, Takeda K. Detection of coronary artery stenosis with whole-heart coronary magnetic resonance angiography. J Am Coll Cardiol. 2006;48:1946–1950. doi: 10.1016/j.jacc.2006.07.055. [DOI] [PubMed] [Google Scholar]
  • 4.Yang Q, Li K, Liu X, Bi X, Liu Z, An J, Zhang A, Jerecic R, Li D. Contrast-enhanced whole-heart coronary magnetic resonance angiography at 3. 0-T a comparative study with x-ray angiography in a single center. J Am Coll Cardiol. 2009;54:69–76. doi: 10.1016/j.jacc.2009.03.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Wang Y, Riederer SJ, Ehman RL. Respiratory motion of the heart: kinematics and the implications for the spatial resolution in coronary imaging. Magn Reson Med. 1995;33:713–719. doi: 10.1002/mrm.1910330517. [DOI] [PubMed] [Google Scholar]
  • 6.Danias PG, McConnell MV, Khasgiwala VC, Chuang ML, Edelman RR, Manning WJ. Prospective navigator correction of image position for coronary MR angiography. Radiology. 1997;203:733–736. doi: 10.1148/radiology.203.3.9169696. [DOI] [PubMed] [Google Scholar]
  • 7.Taylor AM, Keegan J, Jhooti P, Firmin DN, Pennell DJ. Calculation of a subject-specific adaptive motion-correction factor for improved real-time navigator echo-gated magnetic resonance coronary angiography. J Cardiovasc Magn Reson. 1999;1:131–138. doi: 10.3109/10976649909080841. [DOI] [PubMed] [Google Scholar]
  • 8.Sachs TS, Meyer CH, Irarrazabal P, Hu BS, Nishimura DG, Macovski A. The diminishing variance algorithm for real-time reduction of motion artifacts in MRI. Magn Reson Med. 1995;34:412–422. doi: 10.1002/mrm.1910340319. [DOI] [PubMed] [Google Scholar]
  • 9.Jhooti P, Gatehouse PD, Keegan J, Bunce NH, Taylor AM, Firmin DN. Phase ordering with automatic window selection (PAWS): a novel motion-resistant technique for 3D coronary imaging. Magn Reson Med. 2000;43:470–480. doi: 10.1002/(sici)1522-2594(200003)43:3<470::aid-mrm20>3.0.co;2-u. [DOI] [PubMed] [Google Scholar]
  • 10.Manke D, Rosch P, Nehrke K, Bornert P, Dossel O. Model evaluation and calibration for prospective respiratory motion correction in coronary MR angiography based on 3-D image registration. IEEE Trans Med Imaging. 2002;21:1132–1141. doi: 10.1109/TMI.2002.804428. [DOI] [PubMed] [Google Scholar]
  • 11.Lai P, Bi X, Jerecic R, Li D. A respiratory self-gating technique with 3D-translation compensation for free-breathing whole-heart coronary MRA. Magn Reson Med. 2009;62:731–738. doi: 10.1002/mrm.22058. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Stehning C, Bornert P, Nehrke K, Eggers H, Stuber M. Free-breathing whole-heart coronary MRA with 3D radial SSFP and self-navigated image reconstruction. Magn Reson Med. 2005;54:476–480. doi: 10.1002/mrm.20557. [DOI] [PubMed] [Google Scholar]
  • 13.Nguyen TD, Nuval A, Mulukutla S, Wang Y. Direct monitoring of coronary artery motion with cardiac fat navigator echoes. Magn Reson Med. 2003;50:235–241. doi: 10.1002/mrm.10550. [DOI] [PubMed] [Google Scholar]
  • 14.Nguyen TD, Spincemaille P, Prince MR, Wang Y. Cardiac fat navigator-gated steady-state free precession 3D magnetic resonance angiography of coronary arteries. Magn Reson Med. 2006;56:210–215. doi: 10.1002/mrm.20938. [DOI] [PubMed] [Google Scholar]
  • 15.Keegan J, Gatehouse PD, Yang GZ, Firmin DN. Non-model-based correction of respiratory motion using beat-to-beat 3D spiral fat-selective imaging. J Magn Reson Imaging. 2007;26:624–629. doi: 10.1002/jmri.20941. [DOI] [PubMed] [Google Scholar]
  • 16.Jackson JI, Meyer CH, Nishimura DG, Macovski A. Selection of a convolution function for Fourier inversion using gridding [computerised tomography application] IEEE Trans Med Imaging. 1991;10:473–478. doi: 10.1109/42.97598. [DOI] [PubMed] [Google Scholar]
  • 17.Luis I, Schroeder W, Ng L, Cates J. The ITK Software Guide. Updated for ITK version 2.4. (2) 2005 and the Insight Software Consortium ( www.itk.org)
  • 18.Wong ST, Roos MS. A strategy for sampling on a sphere applied to 3D selective RF pulse design. Magn Reson Med. 1994;32:778–784. doi: 10.1002/mrm.1910320614. [DOI] [PubMed] [Google Scholar]
  • 19.Scheffler K, Hennig J. Eddy current optimized phase encoding schemes to reduce artifacts in balanced SSFP imaging. Proceedings of the 11th Annual Meeting of ISMRM; Toronto, Canada. 2003. p. 294. [Google Scholar]
  • 20.Deshpande VS, Chung YC, Zhang Q, Shea SM, Li D. Reduction of transient signal oscillations in true-FISP using a linear flip angle series magnetization preparation. Magn Reson Med. 2003;49:151–157. doi: 10.1002/mrm.10337. [DOI] [PubMed] [Google Scholar]
  • 21.Li D, Carr JC, Shea SM, Zheng J, Deshpande VS, Wielopolski PA, Finn JP. Coronary arteries: magnetization-prepared contrast-enhanced three-dimensional volume-targeted breath-hold MR angiography. Radiology. 2001;219:270–277. doi: 10.1148/radiology.219.1.r01ap37270. [DOI] [PubMed] [Google Scholar]
  • 22.Aharon S, Oksuz O, Lorenz C. Simultaneous projection of multibranched vessels with their surroundings on a single image from coronary MRA. Proceedings of the 14th Annual Meeting of ISMRM; Seattle, WA, USA. 2006. p. 365. [Google Scholar]
  • 23.Glover GH, Pauly JM. Projection reconstruction techniques for reduction of motion effects in MRI. Magn Reson Med. 1992;28:275–289. doi: 10.1002/mrm.1910280209. [DOI] [PubMed] [Google Scholar]
  • 24.Schaffter T, Rasche V, Carlsen IC. Motion compensated projection reconstruction. Magn Reson Med. 1999;41:954–963. doi: 10.1002/(sici)1522-2594(199905)41:5<954::aid-mrm15>3.0.co;2-j. [DOI] [PubMed] [Google Scholar]
  • 25.Kellman P, Chefd’hotel C, Lorenz CH, Mancini C, Arai AE, McVeigh ER. Fully automatic, retrospective enhancement of real-time acquired cardiac cine MR images using image-based navigators and respiratory motion-corrected averaging. Magn Reson Med. 2008;59:771–778. doi: 10.1002/mrm.21509. [DOI] [PubMed] [Google Scholar]
  • 26.McLeish K, Kozerke S, Crum WR, Hill DL. Free-breathing radial acquisitions of the heart. Magn Reson Med. 2004;52:1127–1135. doi: 10.1002/mrm.20252. [DOI] [PubMed] [Google Scholar]
  • 27.Manke D, Nehrke K, Bornert P. Novel prospective respiratory motion correction approach for free-breathing coronary MR angiography using a patient-adapted affine motion model. Magn Reson Med. 2003;50:122–131. doi: 10.1002/mrm.10483. [DOI] [PubMed] [Google Scholar]
  • 28.Spuentrup E, Katoh M, Buecker A, Manning WJ, Schaeffter T, Nguyen TH, Kuhl HP, Stuber M, Botnar RM, Gunther RW. Free-breathing 3D steady-state free precession coronary MR angiography with radial k-space sampling: comparison with cartesian k-space sampling and cartesian gradient-echo coronary MR angiography— pilot study. Radiology. 2004;231:581–586. doi: 10.1148/radiol.2312030451. [DOI] [PubMed] [Google Scholar]
  • 29.Pipe JG, Menon P. Sampling density compensation in MRI: rationale and an iterative numerical solution. Magn Reson Med. 1999;41:179–186. doi: 10.1002/(sici)1522-2594(199901)41:1<179::aid-mrm25>3.0.co;2-v. [DOI] [PubMed] [Google Scholar]

RESOURCES