IDEAL Water-Fat Decomposition with Multipeak Fat Spectrum Modeling

Pp t f j p f w e e t s p ψ π π α ρ ρ 2 ) ( ⋅ ⋅ + = = (relative to water ) and the relative amplitude of the p th fat peak (p=1, …, P), respectively, with Σαp=1. If the fat spectrum is known (Δfp and αp), water, fat and the Bo inhomogeneity map (ψ) can be estimated by using the conventional IDEAL algorithm [5] modified to replace the single exponential fat-associated signal weighting ( t f j e 1 2 Δ π ) with a weighted sum of exponentials ( Δ ⋅ t f j p p e π α 2 ). The values of Δfp are known to be relatively constant, so they are considered known and the values labeled in Figure 1 are used. Because αp may change slightly from scan to scan, as a result of differences in relaxation parameters between peaks, it is directly estimated from the 3pt data. Due to the limited temporal sampling with the 3-pt data, we only estimate the relative amplitudes of the 3 primary peaks in the fat spectrum (peak 1-3 in Figure 1). Peak 4 also has appreciable signal; however it is very close to peak 1, so the phase evolution between the two peaks is small (~π/6) with the 3-pt IDEAL echo times. Peaks 5 and 6 are small, and thus can be ignored. We first perform a conventional 3-pt IDEAL reconstruction, from which the pure fat pixels are automatically selected for the spectrum calibration and their fieldmap values are obtained. The identification of fat pixels can be achieved in an automatic fashion by comparing the water and fat contents in a pixel. At these fat pixels, the field map is demodulated from acquired signals. The relative amplitudes can thus be estimated independently at these pixels using a linear least squares inverse by treating the signal intensities at the 3 peaks as three independent species. Finally, αp estimated at these fat pixels are averaged to obtain the final relative amplitudes. In practice, this spectrum self-calibration processing is performed at the center slice of a dataset and the calibrated αp are used at all slices. Phantom, volunteer and patient scans were performed on GE 1.5T TwinSpeed and 3.0T VH/i (HDx, GE Healthcare, Waukesha, WI) scanners with informed consent and permission from our Institutional Review Board (IRB) for all human scanning. Images were collected using a fast spin-echo (FSE) sequence and a 3D spoiled gradient echo (SPGR) sequence modified for use with the IDEAL method. The echo times optimized for conventional 3-pt IDEAL were used [7]. Results Results from over 100 datasets were obtained in a variety of clinical applications to demonstrate the improved water-fat decomposition using the MPIDEAL, including knee, ankle, breast, spine, brachial plexus, pelvis and abdomen. Representative results from four scans are shown in Figure 2. While conventional IDEAL achieves uniform water-fat separation, residual fat signal in the water images is evident as a result of the fat side peaks. Fat is better suppressed in the MPIDEAL water images, where cartilage is better depicted (in knee and ankle) and the contrast between the muscle and the surrounding fatty tissues is greatly improved. For a given dataset, fat appears uniformly dark at all slices from MP-IDEAL, supporting the assumption that the fat pixels within the same dataset can be characterized by the same spectrum. The excellent results also suggest that the spectrum self-calibration algorithm provides sufficiently accurate estimates of αp. Discussion and Conclusion We have modified the signal model for the IDEAL water-fat separation method, in order to create a more accurate representation of the fat signal evolution. Our results demonstrate great improvement in image quality with decreased fat signal in the water images. The signal model used in MPIDEAL requires accurate knowledge of the fat spectrum. With the spectrum self-calibration method, MP-IDEAL’s sensitivity to potential spectrum variation is reduced and no additional scanning is required. The selfcalibration method assumes that all fat pixels in the dataset follow the same signal behavior. One limitation is that it doesn’t take into consideration the possible intra-data spectrum variation in different disease states or fatty tissues, although this effect has not been observed from our experiments. Nonetheless, the modified multipeak model is more representative of the true fat spectrum than the previous single peak model. In conclusion, MP-IDEAL provides greatly improved fat suppression, which may have useful applications for fat quantification.