Table 9.

Comparative analysis of SR Models.

Ref.	Year	Model	Main Features	Unsolved Challenges
[153]	2018	DeepResolve	Offered the benefit of generating high-resolution thin-slice images while reducing scan time	Focused on magnitude data instead of complex or multichannel data, which may limit the output fidelity
[165]	2018	DDCN	Improved resolution through dense connections, efficient parameter sharing, reduced overfitting	-Increased computational complexity due to dense connections, potential overfitting, sensitivity to network architecture and hyperparameters
[154]	2019	SMORE	Enhanced edges without creating artificial structures and improved both visual and quantitative metrics	Does not address motion artifacts and requires accurate knowledge of the point spread function
[155]	2019	FSCWN	Captured and preserved fine details for better reconstruction using fixed skip connections	Limited generalization to different imaging settings and clinical applicability
[156]	2020	DELNet	Enhanced SR through ensemble learning, leveraging complementary priors	Increased computational complexity due to ensemble size and dependence on diverse ensemble members
[157]	2020	SRNet & UNet	Improved image quality and spatial details in cardiac MRI scans, with the potential for reduced scan time and increased temporal resolution	Lack of a reference standard for accurate comparison, along with limited clinical evaluation and a small patient sample size
[158]	2020	4DFlowNet	Achieved an upsampling factor of 2 and effectively reduced noise in the images	Increased computational complexity and dependence on accurate flow dynamics modeling
[159]	2021	3D UNet	Improved SR of dynamic MRI, fine-tuning for specific applications	Increased computational complexity due to fine-tuning and potential overfitting
[166]	2021	VDR-net	Achieved better resolution of reconstructed MRI images through a Very Deep Residual network (VDR-net) and 2D Stationary Wavelet Transform	Focused on single-image super-resolution and may not have been directly applicable to multi-frame or dynamic imaging scenarios
[160]	2022	DC-CNN	Enhanced the quality of MRIs without relying on raw k-space data	Sensitivity to training data quality and limited interpretability of the learned features
[161]	2022	SRflow	Achieved enhanced spatiotemporal vector field resolution, resulting in more precise quantification of hemodynamics	Generalizability to different datasets and anatomical regions, potential information loss or artifacts during SR, and the complexity of learning vector-field data
[162]	2022	DEGRNet	Utilized clinical image resources without specific HR training images, making it compatible with diverse medical imaging modalities	Limited to 2D super-resolution and potential computational overhead from iterative back projection method
[163]	2022	3D CNN	Clinical assessment of brain SR, improved image quality, accurate structural details	Does not focus on smaller and more subtle lesions especially smaller lesions.
[164]	2023	PFRN	Performed feature extraction directly on LR-MRIs while retaining a significant amount of feature information, enabling the extraction of HF details during the reconstruction process	Assessment on diverse clinical CMRI data is needed to validate PFRN’s generalizability
[167]	2023	CycleGAN	Addressed the limitations of non-blind approaches by utilizing a CycleGAN-based model for domain correction and an upscaling network for reconstruction	Lack of evaluation on clinical datasets