CINENet: deep learning-based 3D cardiac CINE MRI reconstruction with multi-coil complex-valued 4D spatio-temporal convolutions

Thomas Küstner,Kerstin Hammernik,Pier Giorgio Masci,René M Botnar,Niccolo Fuin,Claudia Prieto,Daniel Rueckert,Reza Hajhosseiny,Haikun Qi,Aurelien Bustin,Radhouene Neji

doi:10.1038/s41598-020-70551-8

Abstract

Cardiac CINE magnetic resonance imaging is the gold-standard for the assessment of cardiac function. Imaging accelerations have shown to enable 3D CINE with left ventricular (LV) coverage in a single breath-hold. However, 3D imaging remains limited to anisotropic resolution and long reconstruction times. Recently deep learning has shown promising results for computationally efficient reconstructions of highly accelerated 2D CINE imaging. In this work, we propose a novel 4D (3D + time) deep learning-based reconstruction network, termed 4D CINENet, for prospectively undersampled 3D Cartesian CINE imaging. CINENet is based on (3 + 1)D complex-valued spatio-temporal convolutions and multi-coil data processing. We trained and evaluated the proposed CINENet on in-house acquired 3D CINE data of 20 healthy subjects and 15 patients with suspected cardiovascular disease. The proposed CINENet network outperforms iterative reconstructions in visual image quality and contrast (+ 67% improvement). We found good agreement in LV function (bias ± 95% confidence) in terms of end-systolic volume (0 ± 3.3 ml), end-diastolic volume (− 0.4 ± 2.0 ml) and ejection fraction (0.1 ± 3.2%) compared to clinical gold-standard 2D CINE, enabling single breath-hold isotropic 3D CINE in less than 10 s scan and ~ 5 s reconstruction time.

Highlights

Cardiac CINE magnetic resonance imaging (MRI) is the gold standard for the assessment of cardiac morphology and function
In terms of computational efficiency, we report an average training duration of ~ 35 min/epoch, i.e. ~ 24 h in total for ~ 40 epochs on 2 GPUs (Nvidia Titan RTX) and we observed an average reconstruction time of ~ 5 s for CINENet and ~ 2 min for a GPU-accelerated compressed sensing (CS)
We have proposed a novel reconstruction method, named CINENet, for 3D cardiac CINE MRI based on a deep learning network which enables highly accelerated imaging sequences

Summary

Introduction

Cardiac CINE magnetic resonance imaging (MRI) is the gold standard for the assessment of cardiac morphology and function. Further images in other long axis orientations are required for a comprehensive assessment of cardiac morphology and function which in turn requires multiple acquisitions to be performed in several geometric views and thereby increasing overall planning and scan time To overcome these limitations, 2 D1,2 and 3D3–5 free-breathing cardiac CINE imaging with retrospective motion correction have been proposed to minimize slice misalignment and improve patient comfort. To increase the LV coverage, reconstruction of pseudo 3D cardiac CINE datasets from multiple multi-slice anisotropic 2D volumes by using motion-corrected superresolution frameworks have been p roposed[10,11] This requires several low-resolution scans (in different orientations) in the order of several minutes and depend on slice-to-volume registration accuracy. Reconstructions can be accomplished with iterative algorithms that use a fixed sparsity-promoting transformation[20] or that adaptively derive the optimal sparse representation from the data themselves, known as dictionary learning[21]

Methods

Results

Conclusion