Accelerating cardiac cine MRI using a deep learning-based ESPIRiT reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Accelerated MRI acquisition is crucial for cardiac cine imaging to minimize motion artifacts.
• Existing deep learning reconstruction methods face limitations, such as field-of-view restrictions.

Purpose of the Study:

• To introduce DL-ESPIRiT, a novel framework combining parallel imaging and deep learning for robust 2D cardiac cine MRI reconstruction.
• To overcome field-of-view limitations in deep learning MRI reconstruction.

Main Methods:

• Developed DL-ESPIRiT, an unrolled neural network using an extended coil sensitivity model.
• Implemented a (2+1)D spatiotemporal convolution design for dynamic MRI.
• Trained the network on fully sampled 2D cardiac cine datasets and compared it with SENSE-ESPIRiT on retrospectively undersampled data.

Main Results:

• The (2+1)D DL-ESPIRiT method achieved significantly higher fidelity reconstructions compared to SENSE-ESPIRiT (P < .001).
• Improved image quality led to more accurate segmentations of left ventricular volumes.
• Demonstrated feasibility on prospectively undersampled datasets.

Conclusions:

• DL-ESPIRiT effectively combines parallel imaging and deep learning priors for high-fidelity reconstruction of accelerated cardiac cine MRI.
• The method addresses FOV limitations and enhances reconstruction accuracy.
• Further validation is needed for prospectively undersampled data.