Myocardial T1 mapping sequences are commonly performed under breath-hold conditions. However, motion between T1-weighted images can be observed in ~50% of acquisitions. Image registration algorithms can be used for motion correction but are not available on all scanners, can occasionally fail and remain challenging using low contrast sequences such as saturation-based T1 mapping techniques. In this study we sought to develop and evaluate a novel motion-insensitive T1 mapping reconstruction approach which automatically discards misaligned/artefact T1-weighted images.
Proposed reconstruction: T1-weighted image selection is achieved using an iterative process. At each iteration, several T1 map are reconstructed using different image subsets (all images excluding one). T1 fitting is performed using an exhaustive search of the minimal error between the normalized measured signal and a dictionary of a normalized 1-parameter model (S(t)=1–2*e(-t/T1)). The mean fitting error of each reconstruction (ME) is computed over the myocardium/blood area (obtained by manual delineation of the epicardial contour on the first image, which serves as reference). An image is discarded when its associated subset (where this image was excluded) leads to substantial ME reduction (ME < Median - 6*median absolute deviation (MAD) of all MEs). This process is then iterated if an image has been discarded. A maximum number of iteration of 3 was used. Final T1 maps are reconstructed using the conventional approach (i.e. 3-parameter model fit with Look-Locker correction5) and the preselected T1-weighted images. This iterative reconstruction process has been implemented on graphic processing unit (GPU) to reduce the computation time.
Experimental validation: All imaging was performed using a 1.5T Magnetom Aera scanner (Siemens Healthcare, Erlangen, Germany). All T1 mapping scans used a MOLLI sequence with bSSFP readout (TR/TE=2.5ms/1.2ms, flip angle=35°, FOV=360×306mm2, voxel size=2.2×2.5mm2, slice thickness=8mm, bandwidth=1085Hz/Px, GRAPPA acceleration factor=2, partial Fourier=0.87, MOLLI scheme: 5-(3)-3). All reconstructions were performed offline using an in-house software.
Study #1: Evaluation of the algorithm using in-vivo data with simulated motion. Five healthy volunteers were recruited and imaged using the MOLLI sequence acquired under breath-hold conditions (note the scan was repeated if the breath-hold was not successful). Multiple datasets were then simulated from each volunteer by introducing a 1D foot-head translation motion (amplitude={2,4,6,8,10} pixels) in one T1-weighted image (thereafter referred to as moved image). Each dataset was reconstructed using the proposed and the conventional (i.e. using all images) approaches. Reference MOLLI T1 maps were also reconstructed from each volunteer using the raw (no motion) MOLLI images and the conventional 3-parameter model fitting. Correct discard rate and T1 bias over the entire myocardium (with respect to reference MOLLI T1 map) were evaluated.
Study #2: Demonstration of proposed approach in the presence of physiological motion. One healthy volunteer was recruited and imaged twice using the MOLLI sequence: once under breath-hold conditions, one under free-breathing conditions. Each scan was reconstructed using the proposed and the conventional MOLLI reconstruction.
Study #1. One misaligned image in the MOLLI series can lead to important average myocardial T1 bias using the conventional reconstruction (>100ms, see Figure 1). The proposed approach could identify the motion-corrupted images in the majority of simulated cases and reduced T1 bias to <10ms in all tested cases. Figure 2 shows representative T1 maps reconstructed from different simulated motion cases from the same dataset using the proposed and conventional approaches. The proposed reconstruction improved the quality of native T1 maps in all cases and led to similar quality as the reference T1 map.
Study 2. The image registration step failed in the free breathing MOLLI acquisition and resulted in T1 bias in few myocardial segments using the conventional reconstruction (highlighted segments in bullseye plot in Figure 3). The proposed reconstruction provided improved T1 map quality similar to the breath-hold MOLLI acquisition.
The computation time of the proposed image selection process was 500ms in the worst case (3 iterations).
[1] Moon JC, et al., Myocardial T1 mapping and extracellular volume quantification. J Cardiovasc Magn Reson 2013;15(1):92
[2] Messroghli DR, et al., Modified Look-Locker inversion recovery (MOLLI) for high-resolution T1 mapping of the heart. Magn Reson Med 2004;52(1):141–146
[3] Xue H, et al., Motion correction for myocardial T1 mapping using image registration with synthetic image estimation. Magn Reson Med, 2012;67(6):1644-1655
[4] Roujol S, et al., Adaptive registration of varying contrast-weighted images for improved tissue characterization (ARCTIC): Application to T1 mapping. Magn Reson Med, 2015; 73(4):1469-1482
[5] Deichmann R, Haase A. Quantification of T1 values by SNAPSHOT-FLASH NMR imaging. J Magn Reson, 1992;96(3):608–612.