An unsupervised deep learning method for multi-coil cine MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Deep learning, particularly Convolutional Neural Networks (CNNs), has shown promise in cardiac MRI reconstruction.
• Existing deep learning methods require large fully sampled datasets, which are difficult to acquire for cardiac MRI.
• The impact of coil correlation on deep learning-based dynamic MR imaging reconstruction remains understudied.

Purpose of the Study:

• To propose an unsupervised deep learning method for multi-coil cine cardiac MRI reconstruction.
• To address the challenges of limited data availability and the influence of coil correlations in dynamic MR imaging.
• To develop an efficient reconstruction technique for cardiac MRI.

Main Methods:

• Utilized a time-interleaved sampling strategy to generate fully encoded reference data by merging adjacent k-space data frames.
• Trained a parallel network for individual coil image reconstruction.
• Employed a CNN to combine coil images, implicitly learning coil correlations.

Main Results:

• The proposed unsupervised method demonstrated improved reconstruction quality compared to traditional methods like k-t FOCUSS, k-t SLR, L+S, and KLR.
• Achieved high-quality cardiac MRI reconstructions with significantly reduced reconstruction time.
• Successfully addressed the limitations of big data requirements and explored coil correlation effects.

Conclusions:

• The unsupervised deep learning approach offers an effective solution for multi-coil cine cardiac MRI reconstruction.
• This method overcomes the need for extensive fully sampled data and accounts for coil correlations.
• The technique provides a faster and more robust alternative for dynamic cardiac MRI analysis.