Improved characterization of infarct heterogeneity from high resolution T1* maps using compressed sensing and temporal PCA with weighted total variation

Background Characterization of infarct heterogeneity can inform therapeutic strategies for arrhythmia management of patients with prior myocardial infarction (MI). Multicontrast late-enhancement (MCLE) [1] images along the signal relaxation curve, acquired in a breath-hold ECGgated scan, offer better visualization of MI than IR-GRE. A T1 * map and steady state image Mss are then used to quantitatively characterize infarct heterogeneity. However, motion and constrained imaging durations typically yield low spatial resolution images and blurry anatomical borders in infarcted regions on a T1 * map. This work explores the feasibility of accelerating the MCLE acquisition using Compressed Sensing and temporal Principal Component Analysis (CS-tPCA) to achieve higher spatial resolution images and T1 * maps, while preserving anatomical edges using weighted total variation (TV) in the reconstruction, to eventually improve infarct heterogeneity characterization.

Improved characterization of infarct heterogeneity from high resolution T 1 * maps using compressed sensing and temporal PCA with weighted total variation Li Zhang 1,2* , Prashant Athavale 3 , Venkat Ramanan 2 , Jennifer Barry 2 , Garry Liu 1,2 , Nilesh R Ghugre 1,2 , Mihaela Pop 1,2 , Graham A Wright 1,2 From 18th Annual SCMR Scientific Sessions Nice, France. 4-7 February 2015 Background Characterization of infarct heterogeneity can inform therapeutic strategies for arrhythmia management of patients with prior myocardial infarction (MI). Multicontrast late-enhancement (MCLE) [1] images along the signal relaxation curve, acquired in a breath-hold ECGgated scan, offer better visualization of MI than IR-GRE. A T 1 * map and steady state image M ss are then used to quantitatively characterize infarct heterogeneity. However, motion and constrained imaging durations typically yield low spatial resolution images and blurry anatomical borders in infarcted regions on a T 1 * map. This work explores the feasibility of accelerating the MCLE acquisition using Compressed Sensing and temporal Principal Component Analysis (CS-tPCA) to achieve higher spatial resolution images and T 1 * maps, while preserving anatomical edges using weighted total variation (TV) in the reconstruction, to eventually improve infarct heterogeneity characterization.

Methods
Reperfused MI was induced in 2 pigs by complete occlusion of the LAD artery for 90 min. At four weeks, the animals were imaged in vivo using MCLE after injecting 0.2 mmol/kg Gadolinium-DTPA. The fully sampled dataset of each slice was acquired, using a four-channel anterior cardiac coil array, over a 24-s breath-hold to achieve an in-plane spatial resolution of 1.25 mm (monitored HR = 92 beats/min), and was then retrospectively undersampled in the outer k-space region to yield a net acceleration factor of 2.67. With PCA performed on a low-rank Casorati matrix formed from the central region of k-t space, the principal components (PC) of the temporal signal evolution were extracted. The weighted TV regularization was applied in the CS framework to reconstruct the PC coefficient maps from undersampled datasets. The MCLE images were then obtained by a coefficient-weighted sum of PCs and used to obtain the T 1 * and M ss maps from a non-linear least squares parameter fitting, which were then used in a fuzzy c-means clustering algorithm for tissue classification. For comparison, an alternative reconstruction, REPCOM [2], was also implemented.

Results
From Fig. 1, the T 1 * map from CS-tPCA with weighted TV presents the most sharply defined edges in heterogeneous infarct regions; the classification result from CS-tPCA with weighted TV is more consistent with that from the fully sampled dataset and the histology image, particularly in terms of the features indicated by the arrows. From Fig. 2, there is good agreement in the heterogeneous infarct size between the fully sampled dataset and CS-tPCA with weighted TV, while REPCOM yields weaker agreement with the fully sampled dataset.

Conclusions
We successfully demonstrated that improving characterization of infarct heterogeneity is feasible in a high-spatialresolution acquisition using compressed sensing and   Funding GE Healthcare and CIHR grant.
Authors' details