Table 2.

A summary of articles that use deep learning approaches for image reconstruction in MRI.

Reference	Brief overview
[37]	A DNN model for image reconstruction from subsampled MRI scans. It can also be used for image denoising and super-resolution. However, not all image properties are explicitly exploited.
[38]	A deep learning framework for MR image reconstruction called AUTOMAP. It is accurate when compared to conventional methods. However, it is computationally intensive.
[39]	From significantly undersampled k-space data, a CNN framework for high-quality cardiovascular MR image reconstruction.
[40]	A model that blends variational model mathematics with deep learning. Standard reconstruction techniques are outperformed by the model. There is further work to be done on several types of error measures.
[23]	A DNN-based technique for MR image reconstruction. In the weighted loss function, smaller weights are assigned to noisy training images.
[41]	For rapid and accurate CS-MRI reconstruction, a deep learning model has been developed. There is still a requirement to comprehend the proposed method's design.
[42]	A framework for reconstructing MR images from k-space data that has been undersampled. The structure is also noise-resistant.
[25]	A method for image reconstruction denoising and data integrity enforcement. Due to a decrease in trainable parameters, it does not need a large amount of training data.
[43]	A deep neural network-based image reconstruction model. The computational difficulty of compressed sensing-based approaches was addressed in the model.
[44]	A complete framework for high-resolution MR reconstruction. From noisy, low-resolution clinical MRI data, good-quality pictures are recreated.