A 2-point Dixon Technique Provides Robust Fat Suppression for Multi-mouse Imaging

Abstract

Purpose

To determine whether Dixon-based fat separation techniques can provide more robust removal of lipid signals from multiple-mouse MRI acquired images than conventional frequency selective chemical saturation techniques.

Materials and Methods

A two-point Dixon technique was implemented using a RARE-based pulse sequence, and techniques for multi-volume fat suppression were evaluated using a 4-element array of volume resonators at 4.7T. Images were acquired of both phantoms and mice.

Results

Fat saturation was achieved on all four channels of the multiple mouse acquisition with the Dixon technique, while failures of fat saturation were found with chemical saturation techniques

Conclusion

This proof of concept study found that Dixon fat separation provided more reliable and homogenous fat suppression than chemical saturation in phantoms and in vivo.

INTRODUCTION

Small animal models of human disease play an important role in the development of new therapies for human diseases by providing a complete, functional biological system in which the effects of novel interventions may be tested and characterized. MRI is often used in assessing small animal models of disease because it yields high resolution images with excellent soft tissue contrast, it is non-invasive, and it does not require the use of ionizing radiation (1). Many imaging studies require scanning a relatively large number of mice, which may result in logistically challenging experiments. Efficiency may be improved by simultaneously imaging multiple subjects. To that end, multiple mouse MRI (MMMRI) techniques have the potential to improve experimental throughput, enabling the scanning of entire groups of animals in the time required to image just one, improving statistical power and reducing the cost of powerful imaging techniques in biomedical research (2–4).

One complication inherent to this technique that limits its breadth of application is the poor linewidths that are observed over multiple volumes (2,3). This leads to frequency differences between volumes of interest and to larger linewidths than in single-volume experiments. Independent transceive frequencies for each volume would mitigate dispersion of the center frequency between channels; however, the capacity to adjust the center frequency on a per channel basis is not a standard feature of small animal MRI systems. Independent shim hardware for each volume would obviate the problem of larger linewidths, but at a substantial expense, which is unjustified if other techniques can adequately operate in the presence of the magnetic field inhomogeneities(2).

Fat is hyperintense in a wide variety of imaging protocols, complicating visual and quantitative interpretation of images. Additionally, the chemical shift artifact is greater at the field strengths typical of small animal imaging than those of clinical MRI, unless sequence bandwidth is proportionally increased at some cost to the signal-to-noise ratio (SNR). A number of techniques have been developed to suppress the signal from fat (5). The most commonly used technique to suppress the signal from fat, chemical shift selective fat saturation(6), is subject to suppression failures because it assumes that the absolute precession frequencies of fat and water are constant and known precisely over the entire imaging volume. Field inhomogeneities may cause inadvertent water suppression, substantially reducing the SNR of the image and potentially rendering it uninterpretable(5,6). Dixon techniques, in contrast, exploit the relative precession frequency between fat and water, and may provide a useful alternative means of fat suppression that is insensitive to the B₀ inhomogeneities observed in MMMRI(7–10).

In this manuscript we describe the implementation and validation of a two-point Dixon technique for performing multi-volume fat suppression in conjunction with MMMRI using a four-element array of volume resonators. A RARE imaging sequence was modified to yield images in which water and fat are in-phase and opposed-phase. Images of water/fat phantoms and mice were acquired using this sequence and were compared with RARE images acquired with traditional chemical saturation fat-suppression. A Dixon reconstruction technique was used to separate fat and water signals from in-phase and opposed-phase images. We demonstrate the ability of Dixon-based fat suppression techniques to provide more reliable fat suppression than conventional chemical saturation based techniques for multiple-animal imaging.

MATERIALS AND METHODS

All measurements were performed using a 4.7 T, 40-cm bore Bruker Biospec MRI system (Bruker Biospin MRI, Billerica, MA, USA). All animals were cared for in accordance with Public Health Services Policy on the Humane Care and Use of Laboratory Animals. All experiments and procedures were approved by our Institutional Animal Care and Use Committee, which is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care International.

A RARE sequence was modified to allow acquisition of images with a variable delay of the read gradient with respect to the spin-echo, leading to incomplete rephasing of the chemical shift effect and thus a delay-dependant phase difference between species (11). Although this choice of pulse sequence implementation slightly lengthens the required echo train duration and can reduce slice coverage, at 4.7T the required time shift to produce a 180° phase shift between fat and water is only 0.7 ms per echo, which does not severely compromise slice coverage (9).

In-phase (IP) and opposed-phase (OP) images were acquired of phantoms and in vivo specimens with a RARE sequence and processed using a two-point Dixon reconstruction to yield water and fat images. Signal intensities of each pixel in the IP and OP images may be expressed as

$$\begin{array}{l}{S}_{IP}=(W+F)e^{i{\phi }_0}\\ S_{OP}=\mid W-F\mid e^{i({\phi }_0+{\phi }_1+n\pi )}=(W-F)e^{i({\phi }_0+{\phi }_1)}\end{array}$$

where W represents the signal magnitude arising from water, F represents the amplitude of the fat signal, n is 0 in water dominant voxels and 1 in fat dominant voxels, and ϕ₀ and ϕ₀+ϕ₁ are the background phases of the IP and OP images, respectively. The additional phase, ϕ₁, is due to changes in susceptibility weighting and the location of k-space. The patial dependence of all terms has been omitted for clarity.

The primary challenge of reconstructing water and fat images in Dixon techniques is to identify the phase terms at each voxel in a manner which is consistent across the entire image to ensure that the resulting images are of pure species. Once the phases are identified, calculating the signals of water and fat from the signal equations is straightforward. The spatial behavior of the phase must be considered in order to avoid artifacts in the resulting images.

Various techniques have been proposed to solve this problem. Common approaches involve using a region-growing approach(12), cellular automata(13), or fitting the phase to a polynomial(8). We chose to use a polynomial-fitting based phase correction technique(14) because of its ease of implementation. The background phase of the opposed phase image, ϕ₀+ϕ_1, was fit to a 3^rd order polynomial and removed. Following the approach of Glover and Schneider, we fit single-variable polynomials for each spatial direction independently along the center of mass of the image. The fit was performed on the spatial gradient of the OP image, in which the phase discontinuities due to shifts in fat/water dominance may be thresholded and discarded(14). Post processing was implemented in Matlab.

Four oil/water phantoms were used to verify the IP/OP RARE sequence and Dixon reconstruction algorithm. Cylindrical 30-mm diameter phantoms comprised of soybean oil and tap water were aligned inside a four-element linear array of volume resonators (2). Automatic adjustments, including center frequency, transmit power, and 1st order shims, were conducted through a single channel of the 4-element array. Excitation power was split equally among all volume elements. Following automatic adjustments and sagittal localizers, in-phase and opposed-phase axial T₁-weighted images were acquired through all channels simultaneously. The acquisition parameters were: effective TE/TR = 13/1000ms, FOV = 3cm × 3cm, acquisition matrix = 256 × 256, echo train length = 8, and number of averages = 1.

To verify the performance of the sequence and reconstruction algorithm for in-vivo multi-animal imaging, four nude mice were scanned using the linear coil array and the modified RARE sequence. The acquisition parameters were: effective TE = 60 ms, TR = 4000 ms, FOV = 3 cm × 3 cm, acquisition matrix = 256×192, echo train length = 8 and number of averages = 2 (i.e., two IP acquisitions and two OP acquisitions). For comparison, chemical saturation fat suppression was applied to an otherwise identical T₂-weighted RARE acquisition (except that the number of averages was increased to 4, to provide consistent acquisition time and SNR).

RESULTS

Water- and fat-only images were reconstructed from IP and OP images of the four oil/water phantoms. The water-only images show complete fat suppression with a clean transition at the oil-water interface, as shown in Figure 1. Minor artifacts were observed at the fat/air interface, where rapid phase changes can challenge polynomial phase correction(11).

Figure 1.

Four simultaneously acquired water-only phantom images; shimming was performed on phantom c). Images from the other three channels (abd) show good fat suppression which is not degraded by the differences in the shimming of the four volumes.

Similarly, water-only and fat-only images from each mouse were reconstructed from IP and OP images of four animals scanned at once. Representative water-only, fat-only, and in-phase images are shown in Figure 2. Uniform suppression is again present in each water-only image.

Figure 2.

In-phase (equivalent to an image without fat suppression; row A), water-only (row B), and fat-only images (row C) from one slice of the Dixon acquisition are shown.

Identical multi-volume acquisitions employing chemical saturation fat suppression show inconsistent fat suppression, as shown in Figure 3, and partial saturation of the water signal(15). Fat suppression quality varies substantially between channels: the second channel shows a few local failures of fat suppression, two others show widespread failures, and the fourth channel shows degraded SNR due to suppression of the water signal. Overall, reliable fat suppression was not achieved with chemical saturation techniques.

Figure 3.

Four images acquired using chemical saturation for fat suppression are shown. Shimming was performed over the mouse shown in 4a. Figures 4b, c, and d show images acquired in the other three channels. Degraded SNR is present in these images due to inadvertent suppression of the water signal caused by frequency differences between the volumes of interest.

Same-slice images from a multi-volume acquisition, as illustrated in Figure 4, highlight the relative performance of the use of chemical saturation and Dixon fat separation/techniques. Comparison of the in-phase image with the separate fat-only and water-only images shows good separation of the two species. Some residual water signal can be seen in the fat-only image; this may be an artifact of the phase correction algorithm. The water-only Dixon image shows uniform fat suppression unlike the chemical saturation image, which has local failures of fat suppression(15).

Figure 4.

The in-phase (a), fat-only (b), chemical saturation (c), and water-only (d) images from a selected mouse are shown.

DISCUSSION

Effective fat suppression can be achieved in a multiple-volume environment by using a Dixon fat suppression technique, even in situations where chemical saturation techniques are inadequate. Minor modifications to a RARE sequence permits acquisition of both IP and OP images, which are separated into water and fat images in the post processing.

The Dixon technique used here was validated in phantom and compared to chemical saturation in sacrificed animals, where it was shown to provide more reliable and consistent suppression of the signal from fat. Dixon fat separation techniques avoid the potentially severe suppression artifacts due to B₀ inhomogeneities and frequency offsets among separate volumes of interest, including inadvertent suppression of the water signal, which can reduce image contrast and SNR. In contrast, the use of both IP and OP images to form water-only images increases SNR due to signal averaging (16). If SNR is sufficient, scan time may be reduced by using parallel imaging (17) or homodyne reconstruction(18).

In conclusion, the use of Dixon fat suppression was found, in this proof-of-concept study, to provide more robust and useful multi-volume fat suppression than conventional chemical saturation techniques were able to achieve. Although the shimming limitations of MMMRI have limited studies that use fat suppression, Dixon techniques are less susceptible to these restrictions. This is an important addition to the growing suite of tools enabling high-throughput MRI for biomedical research involving small animal models of disease, and widens the spectrum of high-throughput protocols for biomedical research.