Purpose: Cardiac CINE MRI is one of the most important tools in clinical cardiac MRI exams. The dynamic images of the beating heart allow the physicians to evaluate the ventricular wall motion and calculate the ejection fraction (EF), an important quantitative biomarker of myocardium function. Conventional cardiac CINE uses a stack of 2D slices to cover the entire heart. Due to the requirement of high spatial/temporal resolution and relatively slow speed of MRI systems, the image acquisition of each 2D slice is distributed into multiple heartbeats and synchronized with the Electrocardiogram (ECG). In addition, to mitigate the imaging artifacts induced by respiratory motion, each 2D slice is acquired during a breath-hold of 10-15s. However, the ECG setup and repeated breath-holds increase the scan complexity. Furthermore, the ECG often has reliability issues in the magnetic field and repeated breath-holds may introduce additional errors due to slice miss-registration.
To address these issues, we propose an ECG-free, cardiac CINE solution that covers the entire heart within a single breath-hold. Our solution is based on highly accelerated real-time imaging technique that is enabled by our recently proposed parallel imaging and deep learning combined (PI-DL) image reconstruction algorithm. In this study, we evaluated the proposed solution in healthy volunteer subjects and compare its performance with images acquired using conventional protocol.

Methods: In our protocol, the 2D k-space matrix (192×120) was under-sampled 5X to achieve 72ms temporal resolution. View-sharing was used to further reduce the temporal resolution to 48ms. ECG trigger is not needed since the images were acquired in real-time without k-space segmentation. The acquisition of 8-10 slices was fitted into a single breath-held in an interleaved fashion and each slice was scanned for 1.5 seconds to ensure complete coverage of a cardiac cycle. The entire protocol takes 15 seconds within a single breath-hold.
The under-sampled k-space were reconstructed using our recently proposed PI-DL method, which is described in a separate submission. The PI-DL algorithm learns the relationship of the under-sampled images/k-space to the full-sampled ones. Multiple PI modules were inserted into the convolutional neural network (CNN) to take advantage of the information provided by multiple receiver coils.
Our study includes 6 healthy volunteers. The scan includes both conventional cardiac CINE and the proposed single breath-held, ECG-free CINE. The fully sampled k-space from conventional CINE was retrospectively under-sampled to form the training dataset, which consists of 1000 training samples from 5 subjects. The under-sampled k-space from the proposed methods on the 6th volunteer was applied to the trained CNN and generate the cardiac CINE images.

Results: The images of the proposed cardiac CINE protocol have comparable resolution and image quality with the ones from conventional protocol, although it was acquired using a single breath-hold and without ECG triggering. The error between the EFs calculated from the two sets of images was less than 5% range.

Discussion: Our study shows the feasibility of the proposed single breath-hold, ECG-free cardiac CINE solution, which significantly reduces the complexity of conventional cardiac CINE protocol without compromising its clinical performance. The PI-DL algorithm demonstrates the promise of CNN in MR image reconstruction applications. Future study is needed to further optimize and validate the proposed solution using more clinical datasets.



Author: Fei Han

Coauthor(s): Ziwu Zhou, B.S., Vahid Ghodrati, B.S., Yu Gao, B.S., Yingli Yang, Ph.D., and Peng Hu, Ph.D.

Status: Work In Progress