|Home | About | Journals | Submit | Contact Us | Français|
A simple technique is implemented for correction of artifacts arising from non-uniform T2-weighting of k-space data in FSE-based PROPELLER (Periodically Rotated Overlapping ParallEL Lines with Enhanced Reconstruction). An additional blade with no phase-encoding gradients is acquired to generate the scaling factor used for the correction. Results from simulations and phantom experiments, as well as in vivo experiments in free-breathing mice, demonstrate the advantages of the proposed method. This technique is developed specifically for high-field imaging applications where T2 decay is rapid.
Fast spin echo (FSE) methods are becoming increasingly popular in MR imaging due to excellent T2-weighted contrast and reduction in imaging time (1). Fast spin echo sequences can be used with a number of different acquisition strategies: such as Cartesian, projection, PROPELLER, spiral (2-4). FSE, when used in non-Cartesian acquisitions has the additional benefit of better motion immunity (5).
One inherent problem with FSE imaging is the non-uniformity in the acquired data arising from T2 decay (6,7). This occurs when there is substantial delay (and hence T2 decay) between the first and the last line of k-space acquired in a single excitation. T2 decay is more rapid at high-field strengths due to higher susceptibility-induced losses (8). Thus, signal decay between consecutive echoes is more prominent. Correction of artifacts associated with this problem is thus more important at high-fields (9).
We recently demonstrated a technique for preclinical T2-weighted imaging at high-fields using FSE-based multi-shot PROPELLER (10). The motion correction ability of PROPELLER acquisition and reconstruction makes this technique ideal for imaging free-breathing mice. High-field imaging satisfies the higher spatial resolution requirement (11). The FSE-based approach enables excellent T2-weighted contrast. Higher signal levels are obtained by using a 2-shot acquisition. But this causes a concomitant increase in the artifact levels associated with signal non-uniformity.
In Cartesian FSE, T2 decay causes image blurring and ghosting, which is seen along the phase-encoding direction. The complicated data acquisition strategy in PROPELLER makes interpreting these artifacts more difficult, particularly when considered in conjunction with other sources of artifacts, including motion, phase errors due to eddy currents, and gradient imperfections.
In this work, we describe a simple method for reducing these artifacts by acquiring one additional reference blade during acquisition. This reference data are used to deconvolve the effect of signal decay and the ensuing image artifacts. We evaluated the performance of our technique in simulations, and in phantom experiments, and have shown the results to be in good agreement. We have also demonstrated the technique in a biological system, i.e. lung tumors in a mouse model. Respiratory motion makes the lung a challenging region to image. This reinforces the need to use the 2-shot PROPELLER technique (10). Since the air spaces in the lung yield no signal, the artifacts, as well as their removal are more discernible. The correction enabled us to detect smaller lesions and follow them in a longitudinal study.
PROPELLER data acquisition uses a Carr-Purcell-Meiboom-Gill (CPMG) echo train where each phase-encode line is acquired by a different echo. Thus, each phase-encode line has its own echo time, which is a multiple of the echo spacing. This results in a k-space with non-uniform T2-weighting along the width of the blade, i.e. the phase-encode direction. The decay along the readout direction can be ignored, particularly if the receiver bandwidth is high. We refer to this T2-weighting function along the phase-encode direction as W(ky).
Here, n is the echo index. Acquired data are then given by
The reconstructed image is
I0 is the ideal image reconstructed from S0, which does not experience non-uniform T2-weighting.
Contrast in an image is predominantly determined by the low-frequency data. To obtain higher T2-weighted contrast, early echoes in the echo train are used to sample the outer lines in the blade, while the later echoes are used for the center of the blade. This results in a T2-weighting function, as shown in Fig. 1c.
In conventional spin echo images, W(ky) is constant (Fig. 1a) and its Fourier transform (FT) is a delta function (Fig. 1b). So, the reconstructed image is the ideal image. For the weighting function shown in Fig. 1c, the acquired image is a convolution of the ideal image with the FT of W(ky). From Fig. 1d, note that such a weighting gives rise to blurring (widening of main peak) and ghosting (side lobes) in each blade and hence considerable artifacts in the complete reconstructed image.
We propose a simple correction for these artifacts. From Eq.  and , deconvolution in the image space or alternatively, division in the k-space (by W(ky)) will effectively reduce the non-uniform nature of the acquired data. This would ideally require a complete T2 map of the data, but a simpler (faster), though less accurate method, is to use an average/global T2 value.
W(ky) is derived from an additional acquisition of a reference blade, which is acquired in the same way as the actual data blade (2-shot in this case), but without phase-encoding gradients. So, each line in the reference blade samples the same ky = 0 line in k-space, but at a different echo time. The signal decay seen along the reference blade is thus representative of the T2 decay in the specimen. The scaling factor W(ky) is an array of scalars extracted from the magnitude of the complex-valued data for each line in the blade, and normalized with the magnitude of the first acquired line in that blade—the first echo in the echo train. Each data blade is then divided by this scaling factor [W(ky)−1] to generate a corrected k-space with uniform weighting in the phase direction. The corrected blade now looks more like Fig. 1a instead of Fig. 1c. This correction is carried out for each blade in all the slices.
All imaging was carried out on a 7T, 210 mm bore, Magnex magnet with a GE EXCITE console on the EPIC Lx12.4 software platform (GE Healthcare, Milwaukee, WI). The system uses shielded gradient coils (Resonance Research Inc., Billerica, MA) with a maximum gradient strength of 770 mT/m and a rise time of 100 μs. High-power amplifiers (Copley Controls, Canton, MA) are used to drive the gradients. The strong gradients with high slew rate, high duty cycle, and high power amplifiers are necessary for achieving short inter-echo spacing that is critical at high-fields to minimize susceptibility-induced losses. The RF coil used for imaging is a 35 mm diameter quadrature transmit/receive volume coil (M2M Imaging Corporation, Cleveland, OH).
The data acquisition follows the PROPELLER trajectory discussed previously (10). To enable higher T2-weighted contrast, we used a 2-shot scheme with the early echoes acquiring the periphery of the blade and the later echoes acquiring the center. The sequence was modified to include two additional excitations for one reference blade data collection. The reference blade acquisition is identical to the actual data acquisition, but with phase-encoding turned off. Image reconstruction is carried out in MATLAB (The Mathworks Inc., Natick, MA), and follows the same procedure described previously (3,10), but with the additional module for T2 decay correction. The correction factor is generated from the reference blade data collected for each slice. The correction factor is then used to scale the actual data for that slice, as explained in the Theory section. Figure 2 outlines the data acquisition and reconstruction process.
The phantom used in this experiment was a 30mm diameter bottle of water doped with copper sulphate (5mM concentration) with a homogeneous T2 of 125 ms chosen to highlight the artifacts. The artifacts and the effects of the correction algorithm were studied for three different imaging parameters with varying echo train lengths (ETL): (1) ETL = 8, # blades = 25; (2) ETL = 12, # blades = 16; (3) ETL = 16, # blades = 12. All data were acquired using the 2-shot PROPELLER technique, where the early echoes are at the outer edges of the blade and the later echoes are at the center.
Simulations were performed on a two-dimensional uniform phantom, similar to that used in the phantom experiments. PROPELLER raw data were generated by periodic rotations of the Fourier transform of the phantom. T2 decay along the width of the blade was modeled by scaling the rotated data as shown in Fig. 1c. T2* decay along the readout direction was ignored. Simulations were carried out to correspond with the imaging parameters chosen in the phantom experiments.
All animal studies were approved by the Duke University Institutional Animal Care and Use Committee. The mouse model of human lung adenocarcinoma was generated using the Cre-loxP system. Mice with compound conditional mutations in oncogenic K-rasG12D and the tumor suppressor p53 (LSL-KrasG12D; p53FL/FL) were infected with Adeno-Cre via intra-nasal instillation (12).
The mice were free-breathing and maintained under anesthesia by isoflurane administered via a nose cone. The respiratory rate was monitored throughout the course of the study (SA Instruments Inc., Edison, NJ, USA). Core body temperature was measured with a rectal temperature probe and was maintained between 36-37°C by blowing warm air into the magnet bore. An integrated animal cradle that addressed the anesthesia delivery, physiological signal monitoring, and animal positioning needs was used for these studies (13).
2-shot PROPELLER with late echo ordering with ETL = 10, # blades = 20, and TE/TR = 68 ms/3s was used for all in vivo experiments. Ungated multi-slice datasets covering the lungs with an in-plane resolution of 117 μm and 1 mm slice thickness (21 slices) were acquired in ~40 minutes.
Figure 3 shows the results from simulations (column 1) and phantom experiments on the 7T magnet (column 3) for three different sets of acquisitions with varying ETL. Non-uniformity along the phase-encode direction due to T2 decay cause rings to appear in the reconstructed image of the uniform circular phantom. As the echo train length increases, the time between the first and last echo acquisition increases and thus the magnitude of the signal decay in the acquired data increases. Figure 3 shows that increasing ETL causes more, but narrower, rings to appear in the reconstructed image: 1 ring for ETL = 8 (Fig. 3a, 3c), 2 for ETL = 12 (Fig. 3e, 3g), and 3 for ETL = 16 (Fig. 3i, 3k). This corresponds with the changes in the FT of the T2-weighting function, W(ky), shown in column 2 (Fig. 3b, 3f, 3j). As the ETL increases, the number of side lobes increases, while their width decreases.
The last column in Fig. 3 shows the reconstructed phantom image after correcting for T2 decay using the technique described in this paper. The corrected images have a more uniform signal profile, and the rings can no longer be seen. Another improvement is the absence of the bright edge, caused by high-pass filtering from the T2-weighting function, which is evident in the uncorrected images.
Figure 4 shows three representative slices from a multi-slice dataset acquired in a free-breathing mouse with multiple lung lesions. The first row (Fig. 4a, 4b, 4c) shows the dataset without the correction, while the second row (Fig. 4d, 4e, 4f) is the same dataset reconstructed with the correction algorithm.
The image reconstruction with the correction algorithm is clearly superior. Each slice in Fig. 4 shows improvement in a distinct feature of the image. Note the signal drop-off (arrows) in the muscles in Fig. 4a, which is absent in Fig. 4d. In Fig. 4e, the arrow points to a small lung lesion (0.1mm3), which is not discernible in Fig. 4b. The third slice (Fig. 4c, 4f) shows the liver vasculature. The corrected image shows improvement in both the background liver tissue, as well as in the blood vessels. Smaller branches in the vasculature can be seen more clearly at the edge of the liver in Fig. 4f than in Fig. 4c.
The results of the longitudinal experiment are shown in Fig. 5 with the uncorrected (shown on the left) and corrected (shown on the right) reconstruction. The image sequence shows the longitudinal growth of the small lesion shown in Fig. 4e (also Fig. 5b). Multi-slice datasets covering the thorax and abdomen were acquired with an interval of 12 days between consecutive time points. The lesion grows from 0.10mm3 on day 1 (Fig. 5b) to 0.38mm3 on day 12 (Fig. 5d) and 0.71mm3 on day 23 (Fig. 5f). Artifacts in the uncorrected image make it difficult to detect the lesion in Fig. 5a. The lesion can be seen in Fig. 5c and 5e, though the ghosting and streaking make extraction of any quantitative information difficult.
T2 signal decay along the length of the echo train causes non-uniform weighting of k-space data in FSE-based PROPELLER acquisitions. As shown in Fig. 1d, convolution with the weighting function results in blurring and ghosting in the reconstruction. In PROPELLER, the problem goes a step further as the complete dataset is obtained by acquiring blades rotated around the center of k-space. This causes the ghosts to also be rotated, resulting in distinct rings in the image for an object with homogeneous T2 (Fig. 3). As the ETL increases, the decay within the blade gets larger, W(ky) gets sharper, the width of the side lobes in the FT of W(ky) decreases, and their number increases (Fig. 3b, 3f, 3j). This increases the number of rings in the reconstructed image. Simulation results, as well as phantom data, demonstrate this phenomenon.
For objects with a broader spectrum of T2 values, the artifacts are harder to interpret. Like all data inconsistencies in non-Cartesian datasets, these too manifest as streaks in the image. This is evident from the in vivo images (Fig. (Fig.44,,5).5). Measurement of the aggregate decay yields a first-order measure of the discontinuities, which can be used in the correction algorithm. One might reasonably ask if this is appropriate for a range of different T2 values. We performed simulations on in vivo liver tumor images (not shown) with T2 decay rates spanning the range of T2 values in the image (with normal liver at the low end and tumors at the high end, T2 values range between 27 ms and 74 ms), and the results demonstrated that the correction is not terribly sensitive to the T2 used.
We have used a lung tumor model to demonstrate the advantages of the 2-shot PROPELLER technique and the necessity for correction of the non-uniform T2-weighting in multi-echo acquisitions at high-fields. The dark air spaces in the lungs make the artifacts (bright streaks) more pronounced. For the same reason, removal of these artifacts is also more evident in the lungs.
The correction technique described here brings about a clear improvement in several areas of the image. Note, for example the signal drop-off in the muscles in Fig. 4a. Later echoes in the echo train sample the center of k-space (for heavier T2-weighting), which brings about this loss in low-frequency information. The scaling in the correction technique helps retrieve this information, as is seen in Fig. 4d. Column 2 in Fig. 4 illustrates the improvement in visualization of high-frequency information in the image (tiny lesion) due to the reduction in streaking artifacts after correction. In Fig. 4f, we can see improvement in both the low-frequency (liver parenchyma) and the high-frequency (peripheral vasculature) components of the image.
This technique caused no measurable decrease in image contrast in the lung tumor images. Similar measurements in liver tumor images showed a slight drop in contrast between tumor and normal liver, though the difference was observed to be within the range of standard biological dispersion. This application-specific impact of the correction technique on image contrast should be considered while extending its use to other parts of the body (specifically, if the background is a short T2 species that gets disproportionately amplified) and the correction decay rate duly optimized.
Signal averaging was carried out to ensure adequate signal in the high-resolution in vivo datasets. The imaging time was set to approximately 40 minutes, representing what we believe to be a reasonable tradeoff between image quality and throughput. The correction technique requires only one additional blade and hence, does not cause a significant increase in the total imaging time.
Preclinical imaging is typically carried out at high-fields to obtain higher spatial-resolution. But at higher field strengths, susceptibility-induced losses are larger. Thus for multi-echo acquisitions, the k-space data have non-uniform T2-weighting in the phase-encode direction. In this work, a simple method is proposed to correct for the artifacts caused due to non-uniform T2-weighting of k-space data in FSE-based PROPELLER at high-fields.
This method has been specifically developed for high-field MR microscopy, where T2 decay is more rapid. In the clinical setting, at lower field strengths, the T2 decay is more gradual and the k-space data are not as discontinuous. Nonetheless, this technique may still be applicable for artifact removal in cases where long echo train lengths are used or non-sequential view ordering is employed for blade filling (14).
The authors thank Bradford Perez (David Kirsch Laboratory, Duke University) for providing the lung tumor model and Sally Zimney for editorial assistance. All work was performed at the Duke Center for In Vivo Microscopy, an NIH/NCRR national Biomedical Technology Research Center (P41 RR005959) and NCI Small Animal Imaging Resource Program (U24 CA092656).