Introduction

T1w imaging is a critical component of brain MRI. At ultra-low field, where signal to noise ratio (SNR) is low, use of inversion recovery sequences to maximise signal differences is critical, and optimal contrast can often occur when some tissues are inverted and other are not. Magnitude reconstruction results in rebound artefacts (very dark pixels at interfaces where signs change) and confusing lighter appearance for magnetization that is actually inverted. These challenges are particularly prevalent for neonatal brain MRI, where there is generally less T1 dispersion due to longer relaxation times of immature brain tissue [1], and variation of T1 with maturity is a key marker to observe.

Another challenge in low SNR regimes is Rician bias in magnitude images, which can obscure subtle differences in low signal areas and causes problems with image combination and in model fitting, such as in relaxometry.

Given this context, there is potential for gains from using a phase-sensitive reconstruction [2] that allows signs to be preserved and avoids both noise bias and rectification artefacts. This approach is applied here in an example adult and neonate to demonstrate its utility.

Method & Materials

Imaging was performed on a portable MRI system operating at 64mT (Hyperfine Swoop, hardware version 1.7, software version 8.5.0) utilising a single transmit/ eight-channel receive head coil. A single adult volunteer was scanned with the manufacturer’s standard “T1w – Grey/White” (TR=1.5s, TI=300ms) and “T1w – Standard” (TR=0.88s, TI=354ms) acquisitions utilising a MR-IR-TSE with the following additional sequence parameters: FOV=220x180x200mm, res=1.6x1.6x5mm, TE=5.84ms, BW=64kHz, TurboFactor=24. A single neonate (gestational age: 33weeks; age at scan: 35weeks) was scanned with multiple MR-IR-TSE sequences with varying TI with the following sequence parameters: FOV=220x180x180mm, res=2x2x2mm, TR=1.5s, TI=[300, 400, 500, 600]ms, TE=7.49ms, BW=64kHz, TurboFactor=48. Standard system reconstructions and also raw data, post electromagnetic interference removal [3], were exported. Separate uncombined images were reconstructed for each receive channel and for even/odd echoes using BART‘s non-uniform FFT [6]. Root sum-of-squares (RSOS) images were created by taking the RSOS across coils for the even and odd echoes separately. These were then registered together using FSL FNIRT [7] and averaged to produce the final RSOS output image.

Phase sensitive reconstructions were performed as follows. First receive maps of each coil (Si) were calculated from the longest TI dataset in both the adult and neonatal acquisitions, dividing the image from each receiver (Pj) with the root sum-of-squares (RSOS) combination and applying a median filter of size (7x7). Then a Roemer optimal combination was performed [8] as given by equation 1.

Equation 1

$$I_{opt}=\frac{1}{{\sum }_i^{}{S}_i^2}{\sum }_j^{}\left(P_j{S}_j^{\cdot }\right)$$

Finally, the even and odd echoes were again aligned using non-rigid registration and averaged to produce the final output (Real Recon in figures 1 and 2).

Results

Figure 1 shows the results of the adult data. The top row shows two example slices with a grayscale chosen with a minimum at zero intensity. The standard reconstruction and real reconstruction have superior SNR, with the RSOS demonstrating high noise levels. The standard and real reconstruction differ in the CSF spaces, where the standard reconstruction exhibits non-zero intensity and within the CSF with a dark (rebound) band indicating signal cancellation due to partial volume effects. This is missing from the real reconstruction. The bottom row shows the same data windows to include the negative values present in the real reconstructions.

Figure 2 demonstrates the three reconstructions applied to the neonatal dataset. The images corresponding to TI=300ms and TI=400ms demonstrate improved T1w contrast without the rebound artifacts present on the magnitude reconstructions.

Discussion

In this abstract we have shown the usefulness of complex data for T1w images at ULF. It provides improved contrast depiction with less confusing intensities where there are sign changes. This approach will likely offer benefits to routine weighted imaging (T1w, T2, diffusion and FLAIR) in addition to quantitative methods where Gaussian noise offers benefits to model-fitting. The real reconstruction will also be of use for volumetry, where a sharper boundary is seen between the cortex and CSF. Furthermore, this approach lends itself to acquisitions which require signal averaging.

This work has used a dedicated sequence to estimate the receive fields, however this may be unnecessary as coil loading effects are likely to be limited at this field strength, so that a pre-determined coil sensitivity maps could feasibly be used for all reconstructions.

Acknowledgements

This work is supported by the Bill and Melinda Gates Foundation, the MRC (Translation Support Award: MR/V036874/1), the Wellcome/EPSRC Centre for Medical Engineering [WT 203148/Z/16/Z], and the Medical Research Council Centre for Neurodevelopmental Disorders [MR/N026063/1].