Potential radiation dose reduction in computed tomography with deep learning reconstruction: a retrospective...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Computed tomography (CT) utilizes ionizing radiation, necessitating dose reduction strategies.
• Iterative reconstruction (IR) algorithms have improved CT image quality and dose efficiency.
• Deep learning reconstruction (DLR) represents a novel approach to CT image reconstruction.

Purpose of the Study:

• To evaluate the efficacy of deep learning reconstruction (DLR) in reducing radiation dose for routine clinical CT scans.
• To compare the image quality achieved by DLR against traditional iterative reconstruction (IR) methods.
• To assess the consistency of DLR performance across diverse CT imaging protocols.

Main Methods:

• A retrospective monocentric study involving 13,060 patients undergoing CT scans.
• Comparison of CT scans reconstructed using DLR (CT-DLR) versus IR (CT-IR) algorithms.
• Image quality assessment through qualitative evaluation and quantitative measurement of signal-to-noise and contrast-to-noise ratios on a subsample of 200 patients.

Main Results:

• An overall radiation dose reduction of approximately 20% was observed with CT-DLR compared to CT-IR.
• Specific dose reductions included 22% for chest CT, 21% for CAP oncology, 20% for lower limb CTA, and 19% for head CT.
• The CT-DLR group demonstrated superior subjective and objective image quality.

Conclusions:

• Deep learning reconstruction (DLR) enables significant radiation dose reduction in CT imaging.
• DLR achieves higher image quality compared to iterative reconstruction (IR) algorithms.
• The consistent performance of DLR across multiple protocols supports its broader clinical adoption for improved patient safety.