SCMR 22nd Annual Scientific Sessions
In cardiac MRI, self-gating can be used to extract the cardiac and respiratory motion directly from the data without the use of external devices, i.e. ECG, or breath-holds. Methods like the Principle Component Analysis (PCA) give promising results [1,2] but often suffer from incomplete motion separation . Here, we propose the use of an adapted Singular Spectrum Analysis (SSA) [4,5] to robustly extract and separate respiratory and cardiac motion in radial single-slice imaging.
We extract the central k-space sample of each radial spoke and is for auto-calibration. We then create a Block-Hankel matrix by sliding a window of size W through time for all channels. We zero-pad this matrix to get a matrix A of size [Nt x (Nc x W)], where Nt is the number of spokes and Nc the number of receive channels. The Singular Value Decomposition A = U S VH yields pairs of Empirical Orthogonal Functions (EOFs) as columns of U, that represent respiratory and cardiac motion. These EOF pairs can be used for robust respiratory and cardiac self-gating (Fig. 1).
We perform a 20 second radial single-slice scan with base resolution 160 and slice thickness 7 mm of a human heart (short-axis view) on a SIEMENS Skyra 3T scanner using 30 channels of a thorax and spine coil. We extract the motion signals using the adapted SSA and PCA for comparison. We bin the data in 30 cardiac and 8 respiratory phases, where we treat inspiration and expiration separately [6,7]. We utilize RING gradient delay correction  and an extended XD-GRASP-like  reconstruction with wavelet regularization in the spatial domain and total variation in cardiac and respiratory dimension.
Fig. 2 A) shows respiratory and cardiac quadrature pairs (blue) and the first two components of the PCA (red). While SSA yields clearly separated motion signals, the results of PCA are not suitable for binning.
Fig. 2 B) shows images of the SSA self-gated reconstruction. Depicted is the end-systole and end-diastole of three of the 8 respiratory states. The cardiac and respiratory phases have been correctly classified which yields a high-quality reconstruction.
In this proof of concept study we introduced a novel and robust self-gating method based on Singular Spectrum Analysis. It outperforms the conventional PCA method, which in practice requires further preprocessing steps like band-pass filtering  and coil-selection [10,11] to achieve reliable signal separation.