Deep Learning CT Image Reconstruction in Clinical Practice.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Imaging

Background:

• Computed tomography (CT) is essential in diagnostic medicine, with filtered back projection (FBP) and iterative reconstruction (IR) as current standards for image reconstruction.
• New artificial intelligence (AI) methods, specifically deep learning (DL), are emerging for CT image reconstruction.
• This review explores DL's principles and implementation in CT reconstruction, alongside current limitations.

Purpose of the Study:

• To review the principles of current CT image reconstruction methods.
• To explain the basic concepts of deep learning (DL) and its application in CT image reconstruction.
• To discuss commercially available DL algorithms and their limitations.

Main Methods:

• Review of existing literature on CT image reconstruction techniques.
• Explanation of deep learning (DL) principles and artificial neural networks.
• Analysis of current commercially available DL algorithms for CT reconstruction.

Main Results:

• Deep learning (DL) algorithms, utilizing trained artificial neural networks, offer a novel approach to CT image reconstruction.
• Commercially available DL algorithms demonstrate reduced image noise and enhanced image quality.
• These improvements may increase diagnostic confidence and allow for potential radiation dose reduction.

Conclusions:

• Deep learning (DL) represents a significant advancement in CT image reconstruction, moving beyond iterative reconstruction (IR).
• Current DL algorithms show promise in improving image quality and potentially reducing radiation exposure.
• Further clinical trials are essential to establish diagnostic superiority across a wide range of pathologies.