Kernel Density Estimation for Characterizing Population-Level CT Dose Variability with Acr DIR Benchmarks
Abstract
Purpose
Population-level CT dose monitoring typically relies on summary statistics and benchmark comparisons, which can obscure distributional features such as multimodality and disproportionate upper tails. This preliminary feasibility study evaluated the utility of kernel density estimation (KDE) to characterize CT dose distributions using three years of ACR Dose Index Registry (DIR) data from a single facility and to identify exam categories warranting deeper quality assurance review.
Methods
A retrospective analysis was performed using three years of adult CT facility-level DIR data. Examinations were mapped to ten standard adult DIR exam categories. For each category and year, CTDIvol and DLP distributions were summarized using conventional statistics (median and selected percentiles) and compared with DIR diagnostic reference levels when available. Univariate KDE was applied to visualize distribution shape, assess multimodality, and evaluate upper-tail behavior. Feasibility metrics included successful category mapping, adequate sample size for stable density estimation, and the ability of KDE visualizations to identify categories with non-unimodal or long-tailed distributions.
Results
A total of 191,830 adult CT examinations across three years were included after category mapping. KDE produced stable density estimates for 9 of 10 categories meeting a prespecified minimum sample size threshold (≥100 studies). Compared with percentile summaries alone, KDE identified 7 categories demonstrating multimodal dose distributions and 2 categories exhibiting persistent upper-tail inflation relative to diagnostic reference levels, enabling rapid visual prioritization of exams for further review.
Conclusion
We demonstrated the feasibility of applying KDE to facility-level ACR DIR data for CT dose distribution monitoring. KDE provided complementary insight beyond conventional summary metrics by revealing multimodality and upper-tail behavior and may serve as a practical screening tool for targeted quality assurance.