Introduction

There is growing contemporary interest in addressing the limited (≈3 mm) through-plane resolution in 2D acquisitions^1-3. Sufficiently improved through-plane resolution in one (e.g. axial) orientation may allow multiplanar reformats in other (sagittal, coronal) orientations, obviating direct acquisition in those orientations. One approach is to acquire overlapped slices and correct for the slice profile, a process we refer to as $$$k_Z$$$-space based multislice (KZM) reconstruction³. We have previously shown that the traditional rectangular (rect) excitation pulse causes zero crossings in the $$$k_Z$$$ passband of interest, preventing accurate recovery of signal at those frequencies. We have subsequently shown in resolution phantoms how alternative excitations (e.g. Gaussian, raised-cosine) preserve signal power at these frequencies⁴. The purpose of this work is to study the resolution and SNR aspects of such pulses in more detail, with the targeted application of prostate T2SE MRI.

Methods

Axial excitation was assumed with slice selection (Z) along the superior/inferior (S/I) direction. Three RF profiles were considered: (i) the experimentally measured 3 mm thick rect; (ii) Gaussian; (iii) raised cosine ($$$rCos = \frac{1}{2}\left(1+cos(\frac{\pi k_Z}{k_C})\right)$$$ for $$$|k_Z| \le k_C$$$, 0 otherwise, $$$k_C$$$ tunable). Gaussian and rCos were tuned to have 3 mm thick excitation, defined as 5% response at $$$k_Z = 1/(3 mm)$$$. Figure 1 shows the $$$k_Z$$$ responses of these profiles, illustrating the zero crossing of the rect and monotonically decreasing behavior of Gaussian and rCos. The effects of slice profile were studied with simulation and experimental measurement.

Performance of the axial through-plane, resolution-enhancing reconstruction was assessed in the sagittal plane. For simulation, a natively acquired sagittal image was used, resampled to the desired S/I resolution to provide the input target signal. Next, the simulation imposed onto the target signal via convolution the (complex) slice excitation profile being studied after which Gaussian noise was added at a prescribed level (scaled relative to signal power) to generate the simulated acquired signal. This was sampled at the prescribed slice sampling rate. The simulated acquired signal is expected to have similar characteristics to a sagittal reformat of a native axial signal acquired utilizing the same slice profile. Next, the simulated acquired signal was passed through the reconstruction process³ to generate the (sagittal) reconstructed signal, which could then be compared to the target signal. Ideally the reconstructed signal matches the target signal with no more than the intentionally added noise as a difference. Results from simulation were also compared with sagittal reformats from axial acquisition with and without the resolution-enhancing reconstruction.

The linear reconstruction uses Tikhonov regularization which has input parameter $$$\lambda$$$ to trade off noise with edge recovery performance. In the simulation the optimum $$$\lambda$$$ was taken as that value providing minimum mean square error (MMSE) in the difference between the target and reconstructed signals.

The three excitation profiles have slightly different (±10%) areas along Z and thus slightly different relative SNRs in the assumed acquired measurements. This was accounted for in the simulation by scaling the added noise to provide a fair comparison between profiles.

Results

Figure 2 shows a native sagittal input image (A), the simulated acquired (B) and reconstructed signal (C) assuming the measured 3mm-thick slice excitation profile and 1 mm slice-to-slice increment, and sagittal reformats of experimentally acquired axial images without (D) and with (E) the resolution-enhancing filter. The acquisition for (D) used the 3 mm thick rect excitation with 1 mm increment assumed for (C). Note the improved S/I resolution provided by the filtering (C vs. B, E vs. D) and similarity in appearance of (E) vs. (C), suggesting the validity of the simulation.

Figure 3 shows the impact of varying $$$\lambda$$$ away from the MMSE-minimizing optimal value.

Table 1 is a summary of MMSE values at optimum values for the three slice profiles for two different patient studies.

Figure 4 shows results from another subject, illustrating subtle but noticeable improved S/I sharpness for rCos (C) vs. measured rect (B) and Gaussian (D) for the same slice thickness. (E) was formed using the same data set as (C) but with $$$\lambda=0.017$$$ (smaller than the MMSE optimum $$$\lambda=0.040$$$), providing improved sharpness as preferred by a collaborating uroradiologist. (F) shows the sagittal reformat formed from experimental axial images acquired with the 3 mm thick rect pulse with 1 mm increment, showing similar overall quality with E.

Discussion

Providing a simulation framework for examining slice excitation profile choice allows rapid evaluation of a range of slice profiles. In this preliminary work rCos demonstrated the ability to provide reduced MMSE vs. Gaussian and measured rect responses for equivalent slice thickness. Further work remains to investigate a larger range of candidate profiles and translate any gains anticipated into actual experimental acquisition. While MMSE is a convenient metric to use for similarity, as shown in the example in Figure 4, it may not be the best proxy for radiologist preference⁵.

Conclusion

A simulation framework for investigating the impact of slice profile selection on the final reconstructed output from an intentionally-overlapped-slices acquisition has been developed. Initial results are visually similar when given similar conditions. Future work includes identifying promising novel slice profiles and testing (through new physical acquisitions) their effectiveness.

Acknowledgements

No acknowledgement found.