4D alignment of bidirectional dynamic MRI sequences

Over the past decade, research in medical imaging and visualization has seen a progression from 2D to 3D and now 4D representation of medical data. 4D (3D +time) representation allows one to not only visualize a 3D structure, but also to study its functional movement. However, due to technological limitations no current acquisition modality acquires data in 4D. Instead, 4D data is created from multiple 3D or 2D acquisitions. In this paper we describe a method that spatio-temporally aligns dynamic 2D MRI sequences acquired in bidirectional, orthogonal planes (sagittal and coronal) for 4D visualization. As a proof of concept, we focus on the assessment of swallowing as a representative application of our work. MRI sequences with unknown temporal offsets are acquired through repeated instances of swallowing. We present a method that identifies common regions of interest (ROI) in the orthogonal sequences and combines image intensity profiles from the orthogonal ROIs, via registration to a fiducial volume, to align the MRI sequences. An interesting consequence of using bidirectional acquisition planes is that video sequences in the same acquisition view (e.g all sagittal sequences) automatically get aligned with respect to each other. The bidirectional alignment results are validated by comparing the sum of squared difference (SSD) values over all possible alignments of the sagittal sequences. Experimentally it is found that a minima in the SSD values corresponds to the alignment computed by our proposed method.