Introduction

The relatively high temporal resolution of EPI has been exploited in numerous MR applications for detecting time-dependent haemodynamic responses or reducing motion artefacts. Despite the clear advantages of EPI, its acquisition scheme often hinders the configuration of a high-resolution protocol with an optimal TE due to the inherent requirement of an increased readout length for a large matrix size. This difficulty can be alleviated with the partial Fourier (PF) technique, which effectively shortens the readout length.¹ As a result, many recent high-resolution EPI applications aiming to depict MR signals with a submillimetre voxel size employ the PF technique.^2-5 However, most of these submillimetre-resolution studies have not thoroughly investigated the impact of the PF technique on the spatial resolution of EPI images. Therefore, this work aims to evaluate the spatial resolution of PF-accelerated EPI with various high-resolution imaging protocols and different PF reconstruction methods. The evaluation was performed with a simulation and data sets from a phantom and an in vivo subject at 7T.

Methods

To effectively evaluate spatial resolution, a cylindrical oil phantom containing a pyramid structure was developed in-house (Fig. 1). The pyramid structure consisted of eight microscope glass cover plates, layered sequentially according to the thickness, i.e. 2, 2, 1, 1, 0.6, 0.6, 0.3 and 0.3 mm. Figure 1c depicts its tomographic view at the centre, obtained from a GRE sequence (0.5 × 0.5 mm²). For the evaluation of the spatial resolution of PF-accelerated EPI, various imaging protocols were configured with a number of matrix sizes and PF factors (Fig. 2). In order to simulate the FWHM (full-width-half-maximum) of each protocol, a point-source with a FWHM of one-pixel in width for a vector size of 384 (i.e. 0.5 mm for a FOV of 192 mm) was defined (Fig. 3a). Data sets from the phantom and a healthy in vivo subject screened with a standard safety procedure were acquired using the above protocols. Here, the following imaging parameters were applied on a Siemens Magnetom Terra 7T scanner: TR = 2500 ms, FA = 90°, FOV = 192 × 192 mm², 1 mm thickness and three-fold parallel imaging.

Results

Figures 3a shows an exemplary reconstructed point-source from the ‘384_PF6’ protocol. For the effective visual inspection of the FWHM from each protocol, an ROI was selected around the half-maximum (the blue square in Fig. 3a), and its magnified view is depicted (Figs. 3b-e). Figure 3b shows results for the first comparison pair (see Fig. 2b; ‘384_PF6’ vs ‘288_PF0’). As shown in its first column, the FWHMs of both protocols were identical when the point-source was simulated without consideration of T₂^*-decaying effect and particular reconstruction of PF data (i.e. zero-padding). Theoretically, this is due to the equivalent number of sampled points for both cases (i.e. 288). When the T₂^*-decaying effect (T₂^*: 33.2 ms)⁶ was considered in the simulation, the FWHMs of both cases become wider, with that of ‘384_PF6’ giving a greater value than that of ‘288_PF0’ (2nd column in Fig. 3b). However, the wider FWHM becomes narrower when the missing frequency components in the PF data were reconstructed with their original values (i.e. perfect PF reconstruction); the FWHM of ‘384_PF6’ is shown to be smaller than that of ‘288_PF0’ (3rd column in Fig. 3b). The results of other comparison pairs are also shown in Figs. 3c-e. Here, the narrower FWHM of ‘384_PF5’ compared to ‘384_PF6’ was due to the reduced T₂^*-decaying effect as a result of a shorter readout (i.e. less image blurring).
Figure 4 shows phantom results where the PF data were reconstructed with the zero-filling, submatrix (a Siemens reconstruction option) and POCS methods.^7,8 For all comparison pairs, the images from the zero-filling method depict the performance most degraded with blurring or ringing artefacts. However, these artefacts are shown to be reduced in the submatrix- or POCS-reconstructed images. Here, an enhanced spatial resolution can be observed in the POCS when compared to the submatrix. For all comparison pairs, it was also visually verified that the submatrix- or POCS-reconstructed images in the top row depict higher spatial resolution than those in the bottom row (see yellow arrow) and are in good agreement with the simulation results. The similar spatial resolution performance was also verified with in vivo data (Fig. 5). Furthermore, the phantom and in vivo results also demonstrate that the use of an intensive PF-factor (i.e. 5/8) with a larger matrix size improves the spatial resolution to be better than the moderate PF-factors (i.e. 6/8 and 7/8) with smaller matrix sizes (i.e. ‘384_PF5’ and ‘288_PF5’ against the ‘288_PF6’ and ‘192_PF7’, respectively).

Discussion and conclusions

The results from the simulation, phantom and in vivo data suggest that the number of actual sampled points, i.e. matrix size × PF-factor mainly determines the spatial resolution of PF-accelerated EPI. This indicates that an intensive PF-factor (i.e. 5/8) may be reliably used when a sufficiently large matrix size is provided, as verified here with the following cases: ‘384_PF5’ vs ‘288_PF6’ and ‘288_PF5’ vs ‘192_PF7’. Here, it was also shown that the spatial resolution can be further improved by employing an advanced PF reconstruction method, i.e. zero-padding < submatrix < POCS.

Acknowledgements

No acknowledgement found.