Resolving Uncertainties of IDEAL Fat-Water Imaging Using Magnetization Transfer Effect
Alexey Samsonov1

1Radiology, University of Wisconsin, Madison, WI, United States


IDEAL fat/water imaging often suffers from estimation errors such as fat/water swaps, which can't be removed even by sophisticated algorithms based on field map smoothness regularization. However, these errors may be minimized by supplying the algorithms with an adequate FM prior, which, however, is not generally available. We propose a new method to improve IDEAL robustness which exploits a phenomenon of absence of magnetization transfer (MT) effect in fat for estimation of sufficiently accurate IDEAL field map prior.


IDEAL (Iterative Decomposition of water and fat with Echo Asymmetry and Least-squares) is an efficient approach for separation of fat and water (F/W) MRI signals1. IDEAL algorithms achieve high accuracy by modeling F/W jointly with the magnetic field inhomogeneity (field map, FM) and, for gradient echo sequences, $$$T_2^*$$$ decay. Yet, IDEAL must contend with inherent uncertainties arising from the competing sources of off-resonance (i.e., fat chemical shift and field inhomogeneity). The associated local minima of the nonlinear objective function, when trapped into, may lead to significant errors (e.g., F/W swaps). To reduce these errors, the algorithms typically exploit FM smoothness assumption2, which may fail in case of significant field inhomogeneities. Theoretically, these errors may be also minimized by supplying the algorithms with an adequate FM prior; however, FM pre-estimation from IDEAL images is again confronted by fat interference. Here, we present a novel method to increase IDEAL robustness which exploits a phenomenon of absence of magnetization transfer (MT) effect in fat to attain fat-insensitive FM pre-estimation.


As demonstrated recently3,4, off-resonance MT saturation does not have a detectable effect on the fat signal when applied far from on-resonance (offset frequency $$$\Delta$$$>2.5 kHz5) to avoid direct saturation. This insensitivity is due to absence of efficient mechanisms to transfer magnetization from fat to either water or tissue macromolecular (protein/membrane phospholipid) protons3. Simultaneously, the pulse attenuates water via MT exchange between saturated tissue macromolecules and water protons. Let’s assume that all parameters affecting fat (e.g., repetition time TR, excitation flip angle) are kept the same between two scans acquired without and with MT saturation. Then, the complex-valued subtraction of corresponding images $$$S_n^{off}$$$, $$$S_n^{on}$$$ at $$$n^{th}$$$ echo time $$$t_n$$$ creates MT-attenuated images without fat signal (Fig. 1):

$$S_{W,n}=S_n^{off}-S_n^{on}=(1-A_{MT})We^{-t_n/T_2^*}e^{i\phi t_n}$$

Here, $$$A_{mt}$$$ is MT attenuation, W is water signal, and $$$\phi$$$ is FM. As the fat signal is eliminated, FM can now be robustly calculated in MT-attenuated tissues from $$$S_{W,n}$$$ using standard phase difference methods6 and approximated in the whole object using interpolation-based estimation.


All experiments were performed on a 3.0T GE MR750 (Waukesha, WI). Full-resolution IDEAL scan was followed by low-resolution MT-IDEAL in which MT pulse was positioned in place of later echoes to maintain same TR=23ms (Fig. 2). Other parameters included: 1) IDEAL: TE=[1.5 3.3 5.0 6.7 8.4 10.2 11.9 13.6]ms, 192x160 matrix; 2) MT-IDEAL: TE=[1.5 3.3 5.0]ms, 8-ms, 600 dgr Fermi-shaped MT pulse, $$$\Delta$$$=3 kHz, 192x16 matrix). MT-IDEAL images were subtracted from resolution-matched IDEAL images to yield $$$S_{W,n}$$$ MT-based FM estimation included calculation of phase difference $$$angle[S_{W,3} \cdot conj(S_{W,2})]$$$ , phase unwrapping6 and nonlinear interpolation/smoothing. (The first echo was not included to avoid its prominent phase errors7). The interpolation-based refinement was implemented as a moving local 2D quadratic fit8 weighted by MT-attenuated image $$$|S_{W,2}|$$$ serving as a measure of confidence of the input FM values. The phase-difference processing was also applied directly to IDEAL data without MT IDEAL subtraction to FM prior without fat correction with proposed fat-free MT-based FM. All FMs were used to initialize both standard (non-regularized) IDEAL1 and IDEAL with FM smoothness regularization2.


Figure 3 demonstrates that initialization of non-regularized IDEAL with the proposed MT-based FM resolves F/W swaps observed for default zero-valued initialization. Figure 4 shows that initializing regularized IDEAL by MT-based FM significantly improves accuracy of F/W separation when field variations are significant. FM obtained without MT IDEAL subtraction does not improve separation confirming that proposed MT-based fat correction is critical to obtain prior FM estimate of sufficient accuracy.


The proposed method improves robustness of IDEAL by enabling FM pre-estimation without fat interference. In scenarios with moderate inhomogeneity, our approach improves F/W separation even without computationally demanding FM smoothness regularization (Fig. 3). As MT-based FM initialization facilitates convergence to the correct minima of the objective function, it may improve F/W estimation in suboptimal imaging conditions such as significant field inhomogeneities (Fig. 4) and non-uniform/non-optimized echo spacing. Our approach leads to moderate scan time increase (~10%). This overhead may be reduced in protocols using GRAPPA9 to 5% and 2.5% for 2x and 4x accelerations, correspondingly, if the extra MT-IDEAL scan is utilized for GRAPPA calibration thereby avoiding acquisition of fully sampled k-space center of IDEAL dataset. Ou­r method relies on existence of MT-sensitive tissues, which are naturally abundant in human body (e.g. brain, muscles, collagen-containing tissues, liver). We found that interpolation/smoothing processing yields FM estimate in MT-insensitive tissues (fat, fluids) sufficiently accurate for efficient initialization of IDEAL algorithms. Alternatively, the MT-based FM prior can be incorporated into probabilistic IDEAL framework10 or can be used for initialization of seed areas in region growing algorithms11.


The work was supported by NIH (R21EB018483). We are thankful to Drs. Diego Hernando and Samir Sharma for useful discussions.


1. Reeder SB, Pineda AR, Wen Z, Shimakawa A, Yu H, Brittain JH, Gold GE, Beaulieu CH, Pelc NJ. Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL): application with fast spin-echo imaging. Magn Reson Med 2005;54(3):636-644.

2. Hernando D, Haldar JP, Sutton BP, Ma J, Kellman P, Liang ZP. Joint estimation of water/fat images and field inhomogeneity map. Magn Reson Med 2008;59(3):571-580.

3. Chen JH, Sambol EB, Decarolis P, O'Connor R, Geha RC, Wu YV, Singer S. High-resolution MAS NMR spectroscopy detection of the spin magnetization exchange by cross-relaxation and chemical exchange in intact cell lines and human tissue specimens. Magn Reson Med 2006;55(6):1246-1256.

4. Chen JH, Le HC, Koutcher JA, Singer S. Fat-free MRI based on magnetization exchange. Magn Reson Med 2010;63(3):713-718.

5. Portnoy S, Stanisz GJ. Modeling pulsed magnetization transfer. Magn Reson Med 2007;58(1):144-155.

6. Jezzard P, Balaban RS. Correction for geometric distortion in echo planar images from B0 field variations. Magn Reson Med 1995;34(1):65-73.

7. Hernando D, Hines CD, Yu H, Reeder SB. Addressing phase errors in fat-water imaging using a mixed magnitude/complex fitting method. Magn Reson Med 2012;67(3):638-644.

8. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42(5):952-962.

9. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med 2002;47(6):1202-1210.

10. Yu H, Reeder SB, Shimakawa A, McKenzie CA, Brittain JH. Robust multipoint water-fat separation using fat likelihood analysis. Magn Reson Med 2012;67(4):1065-1076.

11. Yu H, Reeder SB, Shimakawa A, Brittain JH, Pelc NJ. Field map estimation with a region growing scheme for iterative 3-point water-fat decomposition. Magn Reson Med 2005;54(4):1032-1039.


Figure 1: Illustration of insensitivity of fat to off-resonance MT saturation ($$$\Delta$$$=3 kHz). Complex-valued subtraction removes fat components and leaves only water signal in tissues attenuated by MT effect.

Figure 2: Pulse sequence design for MT-based field map estimation.

Figure 3: Effect of our MT-based field map (FM) prior (a) on performance of non-regularized IDEAL in brain. Water and fat components from standard IDEAL (b, c, respectively) shows swaps and errors (white arrows). IDEAL initialized by proposed MT-based FM produces improved water/fat separation (d, e).

Figure 4: Performance of regularized IDEAL with several field map (FM) priors (strong field inhomogeneity, ankle dataset). a,b: FM priors obtained with and without MT IDEAL subtraction. c: Separation results. MT-based FM initialization avoids errors visible in IDEAL initialized by either standard (zero) FM or FM without MT IDEAL subtraction.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)