Model-based alignment of Look-Locker MRI sequences for calibrated myocardical scar tissue quantification

The characterization of myocardial scar tissue in Late Gadolinium Enhancement (LGE) MRI volumes is hampered by the nonquantitative nature of MRI image intensities. Using the widely available Look-Locker (LL) sequence that images the heart using different inversion times a T1 map can be created per patient to calibrate the LGE datasets. However, due to the nature of the LL acquisition, the myocardium is imaged at different phases of the cardiac cycle, resulting in deformations between slices of the LL stack and preventing accurate T1 map estimates. In this paper a method is proposed for the non-rigid alignment of the LL stack that uses a model of the exponential contrast development throughout the LL stack to concurrently align all LL stack slices. The model based alignment is shown to be more robust than a pairwise mutual information based alignment. More importantly, correlations between the relaxivity (R1) map and the LGE intensities (needed for the LGE calibration) are higher using the proposed alignment than when using manual annotations. The model based alignment thereby allows the use of the LL sequence for LGE calibration without manually annotating the (typically) 33 slices in this sequence. After alignment, the myocardium only needs to be annotated in the LGE slice. The latter is also needed for non-calibrated scar quantification and thus requires no additional user effort.