CT image-based evaluation of phantom-derived DLP-to-effective dose conversion factors for adult abdomen-pelvis CT

Effective dose is used in computed tomography (CT) as a population-level quantity for radiation protection, dose monitoring, and clinical studies. It is commonly estimated by multiplying scanner-reported dose-length product (DLP) by a conversion coefficient, or k-factor, because organ doses are not available from routine examinations. Most CT k-factors were derived from computational human phantoms, but phantom-derived estimates have not been systematically compared with effective doses calculated directly from CT images across adults with diverse body habitus. We calculated k-factors using a set of adult abdominal CT images combined with Deep Learning-based automatic organ segmentation and Monte Carlo dose calculation methods and compared them with those calculated from two phantom-derived k-factor approaches: the fixed AAPM TG23 coefficient and body size-dependent k-factors calculated using National Cancer Institute dosimetry system for CT (NCICT). Abdomen-pelvis CT images from 38 adults (18 females, 20 males) spanning water-equivalent diameter (WED) values of 21-41 cm were segmented using TotalSegmentator and an in-house gonad model, converted into voxelized anatomical models, and coupled with a validated CT scanner Monte Carlo model. CT image-based k-factors were calculated as effective dose divided by DLP and evaluated as functions of WED and sex. Image-based k-factors decreased with increasing WED for both sexes, and female values were higher than male values at comparable WED. Over the overlapping WED range, NCICT-based k-factors showed good agreement with CT-based values with the same decreasing trend. The fixed TG23 coefficient introduced systematic bias at smaller and larger body sizes. However, the selected NCICT adult phantoms did not cover the lower-WED end of the CT cohort. These findings support the performance of phantom-based effective dose calculations and indicate the need to extend phantom libraries while continuing CT image-based evaluation across pediatric CT and broader imaging protocols.