Radiation dose optimization potential of deep learning-based reconstruction for multiphase hepatic CT: A clinical and...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medical Diagnostics
• Radiation Dose Optimization

Background:

• Multiphase hepatic computed tomography (CT) is crucial for diagnosing liver conditions.
• Traditional hybrid iterative reconstruction (HIR) methods balance image quality and radiation dose.
• Deep learning-based reconstruction (DLR) offers potential for improved image quality at lower doses.

Purpose of the Study:

• To evaluate the impact of DLR on image quality in multiphase hepatic CT.
• To assess the radiation dose optimization capability of DLR compared to HIR.
• To compare objective and subjective image quality metrics between DLR and HIR.

Main Methods:

• Task-based image quality assessment using a physical phantom to compute contrast detectability.
• Retrospective analysis of 73 patients undergoing multiphase hepatic CT with standard-dose HIR and lower-dose HIR/DLR protocols.
• Comparison of size-specific dose estimates (SSDE), image noise, contrast-to-noise ratio (CNR), and subjective radiologist ratings.

Main Results:

• Phantom studies showed DLR achieved higher contrast detectability at significantly lower radiation doses than HIR.
• Clinical DLR images (lower dose) exhibited reduced noise, higher CNR, and improved subjective quality versus standard-dose HIR.
• Lower-dose DLR achieved superior image quality metrics despite a 52.8% reduction in SSDE compared to standard-dose HIR.

Conclusions:

• Deep learning-based reconstruction substantially improves both objective and subjective image quality for multiphase hepatic CT.
• DLR enables significant radiation dose reduction while maintaining or enhancing image quality compared to HIR.
• DLR represents a promising advancement for safer and more effective multiphase hepatic CT imaging.