Clinical consequences of deep learning image reconstruction at CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Iterative reconstruction techniques are standard in CT imaging.
• Deep learning reconstruction (DLR) presents potential advantages over iterative methods.
• DLR may enable significant radiation dose reduction in CT scans.

Purpose of the Study:

• To evaluate the performance and potential of deep learning reconstruction (DLR) in abdominal CT imaging.
• To discuss the impact of DLR framework, training data, patient size, and clinical task on dose reduction.
• To identify current challenges and future research directions for DLR in CT.

Main Methods:

• Review of current literature on deep learning reconstruction in CT imaging.
• Analysis of DLR performance metrics, including image noise, spatial resolution, and artifact reduction.
• Focus on abdominal applications of DLR.

Main Results:

• DLR decreases image noise without altering noise texture and is less limited by spatial resolution at low doses.
• DLR performance is influenced by the chosen framework, training data, patient size, and clinical task.
• Potential for blurring and artifacts exists with DLR, but it also shows promise in artifact reduction, especially for metal artifacts.

Conclusions:

• DLR offers significant advantages for CT imaging, potentially allowing for more aggressive dose reduction.
• Continued research is essential to fully understand and optimize DLR across diverse clinical scenarios.
• DLR shows promise for abdominal imaging, artifact reduction, and improved diagnostic accuracy with reduced radiation exposure.