Silicone-specific imaging using an inversion-recovery-prepared fast three-point Dixon technique




To demonstrate a new hybrid magnetic resonance imaging (MRI) technique capable of simultaneously generating water-specific and silicone-specific images in a single acquisition.

Materials and Methods

This technique combines short TI inversion-recovery (STIR) technique for robust fat suppression with an efficient fast spin-echo-based three-point Dixon technique for robust separation of remaining water and silicone in the presence of field inhomogeneities. Images demonstrating the feasibility of the technique were acquired with a 1.5-Tesla scanner in a phantom and in a volunteer with both saline and silicone implants in vivo.


The new technique provided water-specific and silicone-specific images of diagnostic quality. Separation of the water and silicone chemical species was complete and satisfactory. Compared with a chemical shift-selective technique, the new technique does not rely heavily on field homogeneity and requires the same or even less scan time to acquire images with similar scan parameters, resolution, and signal-to-noise ratio (SNR).


The feasibility and potential application of the new technique were demonstrated via imaging a phantom and a silicone breast prosthesis in vivo, and it may be used for more consistent imaging of the silicone implants without compromising the image quality or overall scan time. J. Magn. Reson. Imaging 2004;19:298–302. © 2004 Wiley-Liss, Inc.

IT IS ESTIMATED that silicone gel-filled implants have been used as prosthetic devices for mammaplasty in nearly two million women (1). Ruptures or leaks of such implants could occur over time or from trauma and may pose serious health issues for these women (2, 3). Diagnosing ruptures or leakage is problematic since palpation, assessment of tenderness, and standard breast imaging techniques such as mammography and sonography do not generally provide conclusive evidence.

Magnetic resonance imaging (MRI) has proven useful in the diagnosis of ruptures or leakage of silicone gel-filled implants (4–7) and, in general, is more sensitive than competing modalities (8). One of the primary reasons for this high sensitivity is because MRI facilitates the acquisition of silicone-specific images in the breast, permitting unequivocal determination of intra- or extracapsular ruptures of silicone-based prostheses.

There are two general categories of methods for generating silicone-specific images using MRI. Frequency-selective methods use an excitation or refocusing pulse centered on the resonance frequency of silicone in conjunction with other techniques to suppress the water and lipid signals. Standard implementations of this technique use a chemical shift presaturation pulse to eliminate the water component of the signal. Since the resonance frequencies of fat and silicone are too close to separate robustly using a chemical shift-based approach, fat suppression is often achieved via the use of a short TI inversion-recovery (STIR) pulse sequence, which takes advantage of the characteristically short longitudinal relaxation time of fat (5). Such techniques often provide clinically useful images, but because of their reliance on chemical shift-selective radio frequency (RF) saturation pulses, they are inherently sensitive to magnetic field inhomogeneities; therefore, the performance of such techniques could be suboptimal or inconsistent under different scan conditions.

The second general category of methods is based on a phase-selective approach. One of these, the multipoint Dixon (MPD) technique, was originally developed for separation of two dominant chemical species that are present in the human body, namely, fat and water. The general scheme behind Dixon techniques is to acquire multiple images, varying the relative phase shifts between the images for the different chemical species in question. Subsequent image processing can be used to recombine the images arithmetically and generate unique images for each chemical species. An appealing advantage of the Dixon techniques is that they can compensate for effects of field inhomogeneities during the image reconstruction process, removing a major source of failure for the frequency selective-based techniques (9). A disadvantage of the Dixon techniques is that multiple acquisitions are required, leading to longer scan times.

Early attempts to adapt the MPD technique to silicone-specific imaging were based on the approximation that the resonance frequency separation between silicone and water is roughly twice that between silicone and fat (6, 10). Additionally, the technique used spin-echo acquisition and therefore required relatively lengthy acquisition times, leading to reduced slice coverage, compromised imaging parameters, and exacerbated motion artifacts. Initial comparisons between the frequency- and phase-selective methods in the literature demonstrate the degradation of diagnostic quality associated with the limitations of this approach (10). Recently, more sophisticated approaches relying on spectral modeling have been introduced (11). While these models make no assumptions about the frequency spectrum, they do require extensive postprocessing involving inversion of potentially unstable matrices. Further, the implementation of the MPD technique is also based on conventional spin-echo technique and therefore also suffers from long acquisition times.

The goal of this research was to demonstrate a robust and efficient silicone-specific imaging technique by combining the use of inversion-recovery fat suppression with a fast spin-echo-based fast three-point Dixon technique. With the use of this technique, both fat suppression and separation between water and silicone become insensitive to field inhomogeneity. The problem of the long scan time associated with MPD acquisitions was mitigated via use of a modified fast spin-echo sequence for acquisition of the Dixon images (12). The result is a faster and potentially more robust acquisition of water-specific and silicone-specific images for clinical evaluation of patients with silicone implants.


Data Acquisition and Image Reconstruction

To adapt the Dixon technique for silicone-specific imaging, we used a preparatory pulse sequence with the STIR to suppress the fat signals before the Dixon data acquisition. Direct application of the Dixon technique to silicone-specific imaging without fat suppression has to assume that the frequency separation between silicone and water proton resonances is a multiple of the silicone- and fat-resonance frequency difference, as stated in Schneider and Chan (6) and Gorczyca et al (10). When this assumption is not met, separation of silicone, water, and fat is expected to be incomplete since the technique is only designed to separate two distinct chemical species. In our technique, this assumption is unnecessary because only two chemical species, water and silicone, are present after fat suppression.

We reduced the long scan time required to acquire the multiple images for the three-point Dixon technique by using a modified fast spin-echo technique for data acquisition. In principle, fast spin echo can be used to increase the scan efficiency relative to that of conventional spin echo by a factor roughly equal to the echo train length selected. However, this ideal increase in scan efficiency was not achieved in earlier implementations of the fast spin-echo-based Dixon technique (13) because the required echo shifts resulted in a corresponding increase in interecho spacing of the fast spin echo. Furthermore, increased interecho spacing exacerbates image quality problems commonly associated with the use of the fast spin-echo technique, such as ghosting and T2 blurring due to the signal modulation along the echo train. To minimize these effects, we implemented a new fast spin-echo Dixon technique in which the required echo shifts are induced using a pair of bipolar gradients around the readout gradients as opposed to time shifting the readout gradients themselves (12). Using this technique, the interecho spacing as well as the fast spin-echo efficiency is preserved for the Dixon acquisition, and the exacerbation of artifacts due to increased echo spacing is minimized.

A natural consequence of using the Dixon acquisition with gradient-induced echo shifts is that the signals are asymmetric for all echoes with nonzero phase shifts between water and silicone. Thus, for phase correction, we used only the central symmetric portion of the data corresponding to the three echoes with shared k-space coordinates (12). Although the images reconstructed from these data are generally of lower resolution than the final images, no negative effect is expected on the final image quality. The underlying reason is that phase errors due to field inhomogeneity and other system imperfections are, in general, spatially slow varying. This fact is often exploited in phase-correction techniques, in which low-resolution phase images are used regularly for increased robustness and image signal-to-noise ratio (SNR) (14).

The phase-correction algorithm used here consisted of a region-growing technique without direct-phase unwrapping followed by low-pass filtering, similar to what is used for Dixon water and fat separation (15, 16). Region growing starts with a randomly selected seed or seeds and proceeds by seeking directional smoothness of a predefined orientation vector field (15). As in the case for water and fat imaging, the orientation vector is defined from two sets of solutions to the signal equations. When the correct solution is chosen, the orientation vector of a given pixel is parallel to the direction of the local field inhomogeneity. With an incorrect choice, the orientation vector will be either +90° or –90° away from the direction of the local field inhomogeneity (17). Once a smooth orientation vector field is obtained by the region-growing and low-pass filtering process, it is used to correct phase errors in the images reconstructed using all the acquired data. After this, simple arithmetic manipulation analogous to the calculations performed in the three-point Dixon technique for separation of water and fat can be used to generate water-specific and silicone-specific images.


All images were acquired on a 1.5-Tesla whole-body scanner (GE Medical Systems, Waukesha, WI). Before the acquisition of the Dixon data, the frequency separation between water and silicone was measured in a phantom to be 240 Hz, in comparison to 215 Hz for the separation between water and fat. We used these values to determine the areas of the bipolar gradients used for echo shifting in the fast spin-echo sequence for both the phantom and human imaging (described below). Three echoes with relative phase shifts of 0°, 90°, and 180° between water and silicone were collected in an interleaved manner by adjusting the area of the bipolar gradients that were played before and after the readout gradients.

The imaging pulse sequence was compiled under the 83M5 software release for GE Signa Lx systems (GE Medical Systems, Waukesha, WI), and the image reconstruction algorithm was implemented in MATLAB (The MathWorks, Natick, MA).

For phantom imaging, we used a bottle of vegetable oil for cooking (Federated Group, Inc., Arlington Heights, IL), a bottle of copper sulfate pentahydrate and sodium chloride solution in distilled water, and a silicone breast implant (Mentor Corp., Santa Barbara, CA) using the pulse sequence and reconstruction algorithm. The same pulse sequence and reconstruction algorithm were also used to image the breast of a volunteer who had undergone radical mastectomy for breast carcinoma followed by mammaplasty using both silicone and saline implants. The phantom and human imaging was performed using, respectively, head and bilateral phased-array breast coils (MRI Devices Corp., Waukesha, WI). We used the following imaging parameters: TR/TE/TI = 3350/68/150 msec, echo train length (ETL) = 12, field of view (FOV) = 18 cm, slice thickness/gap = 4 mm/1 mm, acquisition matrix = 256 × 192, and receiver bandwidth = 16 kHz. A single acquisition of 20 slices took 5:35 minutes of scan time.


Application of this new technique in the case of the phantom exhibited its ability to produce high-quality water-specific and silicone-specific images. Figure 1a–c show the phantom image prior to Dixon processing, the processed water-specific image, and the processed silicone-specific image, respectively. The fat signals, which usually are hyperintense in a regular fast spin-echo image, were clearly suppressed in these images by the STIR technique. With the three-point Dixon technique, clean separation between the remaining water and silicone signals was achieved. It is noted that the residual fat signal (arrows) after STIR fat suppression appears largely on the silicone-specific image because the resonance frequency of fat is close to that of silicone.

Figure 1.

a: One of the unprocessed images from the hybrid STIR and fast three-point Dixon technique of a phantom consisting of a bottle of water solution, a bottle of vegetable oil, and a silicone breast implant. b: The processed water-specific image. c: The processed silicone-specific image. The separation between water and fat signals is achieved with the fast three-point Dixon technique. The residual fat signal (arrows), which is largely made visible at the selected window levels, appears on the silicone-specific image because the resonance frequency of fat is close to that of silicone.

The in vivo imaging of the volunteer who had had both silicone and saline implants placed after mammaplasty confirmed the ability of this new technique to provide excellent water-specific and silicone-specific images. Figure 2a and b, respectively, show the water-specific and silicone-specific images of the woman at a location where only the silicone implant is situated. The two corresponding images at a different location, where both saline and silicone implants are situated, are shown in Fig. 3a and b. The effectiveness of the new technique in suppressing fat and separating water and silicone images is very satisfactory. It was interesting that the saline implant appears correctly in the water-specific image because of its identical resonance frequency with physiologic water.

Figure 2.

The processed images of a volunteer with double breast implants (silicone and saline) at a location where only the silicone implant is seen. a: The water-specific image shows only the breast tissue. b: The silicone-specific image shows an essentially intact silicone implant. As in the phantom, the low signal posterior to the implant that is visible at the selected window level in the silicone-specific image is believed to be from residual fat signal instead of implant leakage.

Figure 3.

The processed images of the same volunteer as in Fig. 2. The images correspond to a location where both silicone and saline implants are seen. a: The saline implant is visible in the water-specific image because saline has the same resonance frequency as water. b: Only the silicone implant appears in the silicone-specific image.


Among all the available imaging modalities, MRI has the unique capability of producing silicone-specific images for evaluating the integrity of silicone implants. Most MRI techniques used previously are either completely (4) or partially (5, 7) frequency selective. Their performance therefore depends on the field homogeneity, leading to potential performance inconsistencies on a patient or scanner basis. By combining inversion recovery for fat suppression with the fast three-point Dixon technique for water and silicone separation, we were able to obtain high-quality silicone-specific images using imaging parameters comparable to those used in conventional breast imaging techniques. This new approach reduces the potential for introduction of additional artifacts, such as motion, that have limited previous attempts to use a Dixon technique for water and silicone separation. The effects of magnetic field inhomogeneity, which can sometimes render the frequency-selective techniques useless for diagnostic purposes, are easily compensated for in this new technique.

The early implementation of the three-point Dixon technique for silicone-specific imaging by Schneider and Chan (6) relied on an approximation of the interrelationship between the resonance frequencies of the three chemical species. Our measurement shows that the frequency separation between water and silicone (240 Hz) deviates significantly from twice the frequency separation between fat and silicone (25 Hz). With fat signal suppressed by STIR before data acquisition, such deviation does not pose any problems for the technique used here. Without fat suppression, however, substantial leakage of water and fat signals into silicone-specific image is expected. Even if the approximation on frequency separations were true, the direct application of the Dixon method on a system with three chemical species would only work when the phase angles of the Dixon acquisition are set to the traditional values of 0° and 180°. Deviation from these values, for example, the 0°-90°-180° acquisition scheme used in this study, would put water and fat out of phase, therefore breaking down the decomposition algorithm used for generating silicone-specific images. This flexibility in choosing the phase offsets is critical for processing reliability of phase correction and for fast spin-echo acquisition timing.

While the three-point Dixon acquisition still requires a total scan time three times that of a comparable scan with a single signal average, the final water-specific or silicone-specific images have an SNR approximately equivalent to that of a regular scan with three signal averages (9, 12, 15). Since most breast MRI exams are done with multiple signal averages, the total scan time of the current Dixon implementation (5–6 minutes for complete breast coverage) is not prohibitively long compared to that used in current clinical practice. Considering that clinical examination of patients with silicone implants frequently requires imaging of both the water and silicone components of the breasts, which entails two separate acquisitions for the frequency-selective approach, the MPD technique actually amounts to a substantial reduction of total scan time.

In cases in which a single signal average is sufficient for SNR, it may be desirable to further reduce the Dixon acquisition time and associated motion artifacts. One potential approach is to trade off the SNR boost afforded by the fast Dixon technique with the increased temporal resolution afforded by partially parallel imaging techniques such as the technique of sensitivity encoding for fast imaging (SENSE) (18). The latter techniques are rapidly becoming standard features on MR scanners, and prior work has already established that the relative phase information between images is unaffected by the SENSE reconstruction process (19). Furthermore, results of initial investigations into the feasibility of using SENSE reconstructions with the fast spin-echo Dixon sequence for fat and water separation have been positive (20). Given this, incorporation of a SENSE-encoding technique to the current fast three-point Dixon acquisition scheme should be rather straightforward.

In conclusion, we developed a silicone-specific imaging technique that is a hybrid of the inversion-recovery preparatory pulse sequence and the fast three-point Dixon technique. This new technique is capable of generating water-specific and silicone-specific images in a single acquisition and in the same or even less scan time with identical scan parameters used by conventional techniques. No limiting assumptions about the interdependence of the resonance spectral distribution of the three chemical species are necessary, and the full benefit of minimized field inhomogeneity effects on the final images resulting from three-point Dixon processing are realized. This new technique is therefore believed to be a more robust and sensitive alternative to current techniques used for MRI evaluation of silicone-based breast implants.