Dose reduction in radiotherapy treatment planning CT via deep learning-based reconstruction: a single‑institution...

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Artificial Intelligence in Medicine

Background:

• Radiotherapy treatment-planning CT (RTCT) requires careful radiation dose management.
• Current dose reduction methods like adaptive iterative dose reduction (AIDR) have limitations.
• Deep learning-based reconstruction (DLR) algorithms offer potential for enhanced image quality and dose reduction.

Purpose of the Study:

• To quantify radiation dose reduction in RTCT using DLR (AiCE) compared to iterative reconstruction (IR; AIDR).
• To evaluate DLR's potential for informing RTCT-specific diagnostic reference levels (DRLs).

Main Methods:

• Retrospective analysis of 4-part RTCT scans (head, head and neck, lung, pelvis) reconstructed with IR (n=820) and DLR (n=854).
• Comparison of 75th-percentile CTDIvol and DLP between IR and DLR reconstructions.
• Calculation of dose reduction rates and statistical significance testing (Mann-Whitney U-test).

Main Results:

• DLR achieved significant CTDIvol reductions of 30.4-75.4% and DLP reductions of 23.1-73.5% across all sites (p < 0.001).
• Greatest dose reductions were observed in head and neck RTCT (CTDIvol: 75.4%; DLP: 73.5%).
• DLR also narrowed dose variability and achieved lower CTDIvol values compared to published national DRLs.

Conclusions:

• DLR substantially lowers radiation dose indices in RTCT.
• The quantitative data from DLR can effectively guide the establishment of RTCT-specific DRLs.
• DLR optimizes clinical workflows by enabling significant dose reduction in radiotherapy planning.