Deep-learning organ segmentation enables anatomy-weighted CTDIvol from archived metadata in multi-vendor study

Organ-specific weighted CT dose index derived from routine CT metadata using deep-learning segmentation can differ from whole-scan CTDIvol by up to 70% within one acquisition, with manufacturer-dependent archive completeness.

• Modulation weights ranged 0.59–1.69, so the organ-weighted index differs from whole-scan CTDIvol by up to 70%.
• Archived CTDIvol metadata survived in only 29 of 40 series, varying by manufacturer; forced reconstruction diverged by 58% and 84% on two scanner models.
• The index is not an absorbed dose; the open implementation does not provide organ absorbed doses.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.