CineVN: Variational network reconstruction for rapid functional cardiac cine MRI

Abstract

Purpose: To develop a reconstruction method for highly accelerated cardiac cine MRI with high spatiotemporal resolution and low temporal blurring, and to demonstrate accurate estimation of ventricular volumes and myocardial strain in healthy subjects and in patients.

Methods: The proposed method, called CineVN, employs a spatiotemporal Variational Network combined with conjugate gradient descent for optimized data consistency and improved image quality. The method is first evaluated on retrospectively undersampled cine MRI data in terms of image quality. Then, prospectively accelerated data are acquired in 18 healthy subjects both segmented over two heartbeats per slice as well as in real time with 1.6 mm isotropic resolution. Ventricular volumes and strain parameters are computed and compared to a compressed sensing reconstruction and to a conventional reference cine MRI acquisition. Lastly, the method is demonstrated in 46 patients and ventricular volumes and strain parameters are evaluated.

Results: CineVN outperformed compressed sensing in image quality metrics on retrospectively undersampled data. Functional parameters and myocardial strain were the most accurate for CineVN compared to two state-of-the-art compressed sensing methods.

Conclusion: Deep learning-based reconstruction using our proposed method enables accurate evaluation of cardiac function in real-time cine MRI with high spatiotemporal resolution. This has the potential to improve cardiac imaging particularly for patients with arrhythmia or impaired breath-hold capability.

Keywords: cardiac imaging; cine MRI; deep learning; image reconstruction; variational network; ventricular function.

FIGURE 1

Architecture of the CineVN consisting of $N=15$ cascades. Each cascade contains a data consistency (DC) and a refinement (R) term. After every third cascade, we insert a conjugate gradient (CG) block. Trainable parameters are denoted in red.

FIGURE 2

Close-ups of exemplary reconstructions of two OCMR datasets retrospectively undersampled with $R\in \{8,20\}$ using the GRO sampling pattern and reconstructed using CineVN and TTV-CS. Red arrows denote areas were improved temporal fidelity is particularly visible.

FIGURE 3

Short-axis slice from a healthy volunteer acquired using the prospective two-shot and real-time protocols, reconstructed using CineVN, Vendor CS, and TTV-CS. Red arrows denote areas were improved temporal fidelity is particularly visible. Yellow arrows show an area with residual aliasing in the CS reconstruction methods and most notably in the TTV-CS reconstruction.

FIGURE 4

Exemplary reconstructions of prospectively acquired real-time cines from two patients. Red arrows denote areas were improved temporal fidelity is particularly visible. Yellow arrows show an artifact from an artificial aortic valve.

FIGURE 5

Evaluation of ventricular function metrics in prospectively scanned healthy volunteers. Left-ventricular ejection fraction (EF), stroke volume (SV), end-diastolic volume (EDV) and end-systolic volume (ESV) are evaluated for each short-axis stack. Peak radial (PRS) and circumferential strain (PCS) are evaluated for mid-ventricular short-axis slices individually. Plots in the left column show the absolute values for the reference GRAPPA 2 protocol and for the accelerated protocols using the proposed CineVN, Vendor CS, and TTV-CS reconstruction methods. Red asterisks indicate a significant difference to the corresponding GRAPPA 2 reference. Bland-Altman analyses are presented for each reconstruction method. Means and limits of agreement in the Bland-Altman plots are combined over the two-shot and real-time protocols.

FIGURE 6

Ventricular function and strain values evaluated on the prospective patient dataset using the proposed CineVN, Vendor CS, and TTV-CS reconstruction methods.