Table 3.

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

Reference	Brief overview
[46]	A model based on Wasserstein generative adversarial networks for 2D CT slice image reconstruction from a small number of prediction images. Expert radiologists must confirm the model's accuracy.
[47]	A U-net-based image reconstruction framework. It is superior to noise and angle artefacts in terms of visual structure preservation, but it is computationally costly and requires large training datasets.
[48]	A more relaxed variant of projected gradient descent (PGD) is used in this model. The results demonstrate that the new technique outperforms the previous one.
[49]	An approach for CT image reconstruction based on deep learning. When compared to other state-of-the-art approaches, the results show enhanced image quality with less image noise.
[50]	A lightweight framework for a few-view CT reconstruction approach. It learns an end-to-end mapping between a few-view picture optimization and a full-view image optimization.
[51]	For high-quality CT reconstructions, a deep learning architecture was developed. The framework is capable of distinguishing and removing noise from the input signal.
[52]	Iterative reconstructions of data from genuine CT systems using a TensorFlow framework. The drawback is that it necessitates the use of graphics processing units (GPUs).
[53]	During reconstruction, a CNN framework is used to remove streaks from CT images. To discriminate between objects and characteristics, the framework requires further training.
[54]	A deep learning model for reconstructing high-quality images from sinogram data. It reduces noise, improves spatial resolution, and is quick without sacrificing quality.
[55]	For CT reconstruction, there is a framework called LEARN. It boosts image quality as well as computational efficiency. The framework still has to be optimized for clinical applications.