There has been increased interest in the quantitative characterization of tissue based on T2. Techniques based on spin-echo (SE) or fast spin-echo (FSE) sequences are time consuming because they require multiple acquisitions for obtaining an adequate number of TE images for accurate T2 mapping. Radial FSE based methods have been introduced for efficient T2 mapping by using TE data from a single k-space data set [1-4]. In fast acquisitions the TE data sets are highly undersampled (4% sampled compared to Nyquist) so specialized algorithms are needed to reconstruct the TE images. One algorithm is based on echo sharing (ES) by retaining data at the center of k-space corresponding to a specific TE while borrowing data from other TEs to fill the outer part of k-space (Figure 1) [1]. Other algorithms used iterative reconstructions where the data is constrained to fit to a pre-determined signal model [2,3].

In breath-hold applications such as abdomen imaging, the need to fit multiple slices and keep the SAR low often requires the use of non 180 degree refocusing pulses in FSE acquisitions. Imperfections in the resultant profiles lead to the generation of indirect echoes which modulate the signal evolution causing a deviation from the ideal single exponential T2 decay model. An alternative formulation, CURLIE, was recently proposed [4] to accurately estimate T2 from radial FSE data even in the presence of refocusing profile imperfections using the slice-resolved extended phase graph (SEPG) model [5]. This iterative method reconstructs TE images by linearizing the non-linear SEPG model using principal component decomposition. The T2 maps are estimated by using a pattern recognition technique [6] that uses a pre-computed dictionary based on the SEPG model (see Figure 1). These T2 estimates have been shown to be more accurate when compared to those obtained using a single exponential model. Due to the iterative nature of the reconstruction algorithm the generation of TE images is computationally expensive. In this work, we explore combining the ES algorithm (for the fast reconstruction of the TE images) with SEPG model fitting (for compensation of indirect echoes) as a fast approach for accurate T2 mapping for radial FSE data.

The efficiency of the ES-SEPG method was quantitatively evaluated using a set of gel phantoms with different T2 values. Data were acquired using a radial FSE pulse sequence on a 3T Siemens scanner with ETL = 32, echo spacing = 7.2ms and 192 radial views with 256 readout points per view. Reference data were acquired using a Cartesian single spin-echo (SE) pulse sequence as this method is not affected by indirect echoes. SE data were acquired with matrix size = 256x256, TR = 1500ms, and it was repeated for different TE values in increments of 7ms. Radial FSE data for 16 subjects were acquired on a 1.5 T Siemens scanner. The imaging parameters were ETL = 32, echo spacing = 6.2ms and 192 radial views with 256 points per view. The range of T2 and B1 values used for the dictionary generation were 20 – 1000ms and 0.7 to 4 respectively. The mean T2 values and standard deviation for the four phantoms estimated by the ES technique (with both the mono exponential fit and the SEPG dictionary fit) and CURLIE are compared to the single-echo spin echo reference in Table 1. Note that the mean T2 values from ES with mono-exponential fitting are overestimated compared to the SE reference due to indirect echoes. The T2 values estimated by CURLIE are similar to the SE reference because the SEPG model used in CURLIE accounts for indirect echoes. When the faster ES is combined with SEPG fitting we obtained T2 estimates as accurate as CURLIE. Representative TE images reconstructed using the ES algorithm along with the T2 map estimated using the SEPG model for a subject with multiple liver lesions are shown in Figure 2. The results from 33 liver neoplasms are shown in Figure 3 and it can be seen that the T2 estimates from ES with the SEPG based fit can discriminate malignancies from benign as well as the more computationally expensive CURLIE algorithm. In this work we present a model based T2 estimation method for data reconstructed using the echo sharing algorithm. The estimates are shown to be better compared to using a single exponential fit and are comparable to those obtained using the iterative CURLIE technique.