Modeling of T2* decay in water/fat imaging: comparison of one-decay and two-decay models

INTRODUCTION Chemical shift-encoded techniques for quantitatively measuring the presence of fat have a number of important applications in MRI, including bone marrow, muscle, brain, liver and heart studies. However, inaccurate modeling of the acquired signal can result in significant bias in the estimates of fat amplitude or fat fraction. Specifically, modeling the T2* decay (or the decay rate R2*=1/T2*) of the signal has been shown to be necessary in order to avoid such bias [1]. In recent works, the possibility of modeling separate decay rates for water and fat has been proposed [2,3]. Even though the two-decay model is more accurate (potentially lower bias), it suffers from increased noise sensitivity (higher standard deviation) with respect to the one-decay model due to the need to estimate an additional nonlinear parameter. In this work, we analyze quantitatively the tradeoff between bias and standard deviation using simulation, phantom and in vivo data. METHODS Phantom construction: A water fat phantom was constructed by mixing water and oil in separate vials with fat fractions (%): 0, 10, 20,30,40,50,60,70,100, as described in [4,5]. Data acquisition: Phantom and in-vivo data were acquired using a spoiled GRE sequence with monopolar readout, 8 TEs with initial TE=1.43ms and TE spacing=2.23ms. The phantom acquisition was performed both with flip angle=8, TR=500ms (giving SNR≈30), and with flip angle==25, TR=2000ms (giving SNR≈90). Each type of acquisition was repeated 128 times to obtain “Monte-Carlo” measurements. Processing: The complex-valued data were processed by fitting (nonlinear least-squares) 0 4 8