Radiation Dose Reduction for 80-kVp Pediatric CT Using Deep Learning-Based Reconstruction: A Clinical and Phantom...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Deep learning-based reconstruction (DLR) shows promise for reducing CT radiation dose.
• Limited research exists comparing low-dose DLR with standard-dose iterative reconstruction (IR) in pediatric CT.
• Application of DLR to low-tube-voltage pediatric scanning requires further investigation.

Purpose of the Study:

• To evaluate DLR's efficacy in reducing radiation dose for low-tube-voltage pediatric CT.
• To compare image quality of DLR with hybrid IR (HIR) and model-based IR (MBIR) at reduced doses.
• To assess DLR's performance against standard-dose IR techniques.

Main Methods:

• Retrospective analysis of contrast-enhanced 80-kVp CT scans in children aged 6 years or younger.
• Comparison of standard-dose HIR with lower-dose HIR, MBIR, and DLR reconstructions.
• Quantitative assessment of image noise, SNR, CNR, and task-based object detection (d') using phantom studies; qualitative assessment by two radiologists.

Main Results:

• Lower-dose protocols achieved a 54% reduction in size-specific dose estimate (SSDE).
• Lower-dose DLR and MBIR demonstrated reduced noise and improved SNR/CNR compared to standard-dose HIR.
• Radiologists rated lower-dose DLR higher than standard-dose HIR for noise texture, sharpness, and overall quality; DLR outperformed HIR and MBIR in phantom object detection.

Conclusions:

• DLR enables substantial radiation dose reduction in low-tube-voltage pediatric CT while preserving or enhancing image quality.
• DLR offers superior image quality and diagnostic performance at lower doses compared to traditional IR algorithms.
• DLR at 80 kVp represents a significant advancement for dose reduction in pediatric CT.