The quantitative assessment of physical tissue properties (qMRI) has great potential for providing disease biomarkers since it facilitates intra- and inter-subject comparison¹. Various acquisition techniques have been proposed to enable its application in clinical practice. However, many of them are 2D methods; most of the 3D qMRI techniques are still very time-consuming and 2D imaging imposes various limitations such as reduced resolution and SNR as well as anisotropic voxel size. Here, we suggest to combine the recently proposed 3D method “Motion-Insensitive Rapid Configuration Relaxometry” (MIRACLE)² with a highly undersampled 3D phase-cycled balanced steady-state free-precession (bSSFP) sequence termed trueCISS³ enabling the acquisition of high-resolution quantitative T1 and T2 maps in less than 4min. Additionally, we use the estimated quantitative maps to synthesize various conventional MRI contrasts as a visual benchmark.

After obtaining written consent, a prototype trueCISS sequence was used to acquire an 8-fold undersampled, 3D phase-cycled bSSFP k-space from one healthy volunteer at 3T (MAGNETOM Prisma, Siemens Healthcare, Germany) using a 20-channel head/neck coil (resolution 1x1x1.3mm3, TR/TE 6.58ms/3.29ms, flip-angle α=15°, 16 phase-cycles with ɸ=0°, 22.5°, 45°, …, 337.5°, acq. matrix 256x256x128, TA 3:31min).

First, a sparse iterative reconstruction was used to reconstruct each phase-cycled image by enforcing data consistency with the acquired data and sparsity in the wavelet-domain⁴. Following the MIRACLE algorithm², the complex phase-cycled 3D images were then discrete-Fourier-transformed along the phase dimension yielding 16 SSFP configurations or “modes”⁵. Subsequently, the lowest-order modes can be used to derive T1 and T2 using a golden section search in each voxel as proposed for TESS relaxometry⁶.

To extract more information, the trueCISS algorithm³ was applied on the phase-cycled bSSFP images as well. To that end, the bSSFP signal model was fitted onto the magnitude of the phase-cycled images and is defined as follows:

$$$M=M_{0}\mid\frac{2\sin\alpha\cos(\frac{\phi-\triangle\phi}{2})}{1+\cos\alpha+\cos(\phi-\triangle\phi)+(4\Lambda-\cos(\phi-\triangle\phi)^{2})\sin(\frac{\alpha}{2})^{2}}\mid$$$

yielding quantitative maps of the equilibrium magnetization M₀, the relaxation time ratio Λ=T1/T2 and the local phase offset due to field inhomogeneities ∆ɸ.

The quantitative maps provided by these two algorithms in conjunction with the forward signal models of different contrast mechanism can be used to generate synthetic images simulating various conventional MRI techniques. Here, we exemplarily used them to generate T2-weighted “Rapid Acquisition with Relaxation Enhancement” (RARE)⁷ and T1-weighted “Magnetization‐Prepared Rapid Gradient‐Echo” (MP-RAGE)⁸ contrasts alongside with a trueCISS image (i.e. an on-resonant bSSFP image) and a maxCISS image (i.e. a bSSFP image with ideal flip angle depending on the relaxation time ratio Λ of a tissue ). A workflow diagram of this prototype algorithm is illustrated in Figure 1.

An axial and sagittal slice of the different quantitative maps M0, T2 and T1 is shown in Figure 2. The quantitative T2 values within regions of interest (ROIs) correspond to what was reported in literature for this measurement method⁹, with white matter ~53ms and grey matter ~88ms. However, the estimate of T1 values is affected by B1-field inhomogeneities resulting in longer values in the centre of the brain. As it has been already discussed for the fully sampled MIRACLE algorithm, the bias in the T1 values can be corrected using an additional B1-field acquisition. However, it appears that T1 values are systematically underestimated potentially due to asymmetries in the bSSFP profile¹⁰.

Currently MIRACLE and trueCISS are applied separately onto the sparse reconstructed phase-cycled images. However, both fitting procedures incorporate relaxation, i.e. T1 and T2 for MIRACLE and Λ=T1/T2 in trueCISS. In the future, both algorithms should be combined by using mutual variables within a single optimization to improve the fitting procedures in order to increase the robustness and quality of the estimates.

The synthetically generated contrasts of the same axial and sagittal slices are shown in Figure 3. All simulated images resemble the contrast of the conventional sequence, their clinical usefulness, however, still has to be validated.

We propose to combine two recently reported techniques: trueCISS and MIRACLE (in a novel undersampled version) to obtain high-resolution three-dimensional quantitative maps of M0, T1 and T2 in clinically feasible times (3:30min in the presented example). The obtained quantitative values were in the expected range in different brain regions and allowed for an application of different forward signal models, yielding simulated RARE and MPRAGE contrasts. The possibility of obtaining high-resolution qMRI images with isotropic voxel size may increase the clinical value of those techniques thanks to the improved conspicuity of small pathologies. Further work should focus on evaluating the clinical usefulness of qMRI as well as synthetic contrasts; furthermore, databases with quantitative values of healthy subjects should be built up, enabling the identification of abnormal tissue from a single-time-point scan.