DeepcomplexMRI: Exploiting deep residual network for fast parallel MR imaging with complex convolution.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Parallel Magnetic Resonance Imaging (MRI) acquisition is crucial for reducing scan times.
• Existing acceleration methods often depend on coil sensitivities or predefined transform priors.
• There is a need for advanced reconstruction techniques that improve image quality and efficiency.

Purpose of the Study:

• To propose DeepcomplexMRI, a novel multi-channel image reconstruction method for accelerated parallel MRI.
• To develop a deep residual convolutional neural network capable of reconstructing high-fidelity multi-channel MR images.
• To evaluate the performance of DeepcomplexMRI against state-of-the-art methods using in vivo datasets.

Main Methods:

• A residual complex convolutional neural network was developed, considering the correlation between real and imaginary parts of MR images.
• The method utilizes a large dataset of multi-channel ground truth images for offline training.
• k-space data consistency was enforced iteratively within the network layers.

Main Results:

• DeepcomplexMRI demonstrated the capability to accurately recover desired multi-channel MR images from accelerated acquisitions.
• Evaluations on in vivo datasets confirmed the method's effectiveness.
• Comparative analysis showed superior reconstruction accuracy compared to existing state-of-the-art techniques.

Conclusions:

• DeepcomplexMRI offers an effective approach for accelerating parallel MRI acquisition.
• The proposed complex convolutional network architecture and training strategy yield high-quality multi-channel image reconstructions.
• This method has the potential to significantly improve the efficiency and diagnostic utility of MRI.