Balanced steady-state free precession (bSSFP) provides high signal in short scan times and useful contrast for visualizing and characterizing many tissues in the body. Despite these benefits, bSSFP scans are susceptible to B0 inhomogeneity, which causes dark bands of signal loss across the images at certain off-resonance frequencies (“banding artifacts”). To mitigate these artifacts, multiple bSSFP acquisitions are often combined, each with a different RF phase increment from TR to TR. This effectively shifts the off-resonance banding artifact position to different locations for each acquisition. These images can then be combined to reduce banding artifact in the final reconstructed image. The individual acquisitions are commonly combined using a sum of squares (SOS) or complex sum (CS) of the constituent images [1]. A new method for combining four or more phase-cycled bSSFP acquisitions was recently proposed by Xiang and Hoff that uses an elliptical signal model (ESM) of the bSSFP signal [2]. It achieves near perfect banding artifact removal under certain conditions. However, an analysis of the SNR performance and banding reduction of the new ESM technique across a range of tissue types, flip angles, and base SNR levels has not yet been performed, nor has the performance of the new technique been quantitatively compared to the SOS and CS methods.

In this study, we follow the statistical analysis framework outlined in [1] to compare the SNR performance and effectiveness at reducing banding artifact of the ESM, CS, and SOS techniques across a range of T1 and T2 values, flip angles, and base SNR levels.

Simulations: A Monte-Carlo simulation was performed as described in [1] to determine how noise propagates when using SOS, CS, and ESM to combine four phase-cycled images. The simulation was performed at flip angles ranging from 10 to 90 degrees in increments of 10 degrees, across a range of tissue T1 and T2 values. In addition to the Monte-Carlo simulations, a simulated phantom image composed of 7 simulated tissues similar to blood, gray matter, white matter, fat, synovial fluid, cartilage, and muscle (T1/T2 = 1500/200, 1800/100, 1000/70, 300/85, 4800/325, 1200/30, 1500/32 ms) was used [3]. A TR/TE of 10/5ms was assumed. Base SNR (as defined in [1]) was varied from 5 to 30 in increments of 5. SNR performance and residual banding artifact were determined from the simulations, again using the techniques outlined in [1].

Actual Data: Four 2D bSSFP phase-cycled acquisitions were acquired (phase increments of 0, 90, 180, and 270 degrees) on a 3T Siemens Trio scanner of both a water/oil phantom and the knee of a normal volunteer. Scan parameters for both scans were: TR/TE=10/5ms, α=30 degrees, FOV=200mm, slice thickness = 5mm, and matrix=256x256. The data was in each case combined with all three techniques (SOS, CS, and ESM), and SNR performance compared with the performance predicted by simulation.

Simulations: The ESM technique achieves similar SNR performance to the CS technique, while the SOS technique resulted in the greatest SNR across all tissues and parameters tested. A summary of these results is shown in Figure 1. When looking at banding artifact reduction, the new ESM technique yields by far the best performance when base SNR is above 30 for most tissues (Figure 2). However, for certain tissues and at low SNR for many tissues, the ESM technique actually performs worse than both the CS and SOS techniques in reducing banding artifact. This can be understood by looking at the shape of the ellipses in the elliptical signal model across a range of tissue T1 and T2 values (Figure 3). When the ellipses become too small the magnitude of the noise is comparable to the size of the ellipses in the complex plane, the signal level estimated by the ESM can vary significantly.

Actual Data: Corrected SNR values for the water/oil phantom can be seen in Table 1 for each of the combination methods. The real data matches the simulation results: SOS has the highest SNR, whereas ESM and CS have similar SNR values. In-vivo knee data can be seen in Figure 4. No residual banding can be seen in any of the methods in this experiment.

Although ESM produces near perfect band removal in high SNR situations, it breaks down for certain tissues and in low SNR situations. In extremely low SNR cases SOS can outperform CS and ESM in banding reduction. In these cases SOS results in the highest SNR and least banding artifact.