Objectives Identify routine diagnostic abdomen CT protocols in adults that achieve lower radiation doses, crowd-sourced from a large CT dose registry. Materials and methods We retrospectively captured acquisition parameters, patient diameter, and patient size-adjusted and unadjusted radiation dose (CTDIvol and DLP). A protocol was defined as a unique combination of facility, scanner make/model, CT indication, and protocol name. We used k-means clustering to classify protocols into clusters based on similarity of average acquisition parameters (mAs, pitch, kV, collimation, scan length, phase count), each representing a distinct pattern of protocol design choices. For each cluster, we summarized the mean technical parameters and dose metrics of its constituent protocols. Results Analyses included 907,992 exams from 1767 protocols at 132 facilities, grouped in 9 clusters. The number of protocols and exams within each cluster ranged from 62 to 508 and 15,317 to 381,457, respectively. Mean size-adjusted DLP varied threefold across clusters, ranging from 486 to 1382 mGy-cm, with no difference in patient diameter. Lowest-dosed clusters 1 and 2 minimized radiation primarily through low kV (around 100). Cluster 8 had the highest acquisition techniques (mean mAs = 338 and kV = 125), and cluster 9 had the highest dose due to phase (mean = 3.3, versus 1.1–1.6 for the remaining clusters). Conclusion The clustering approach offers a new framework for identifying optimized protocols, while highlighting variation in practice and worst-in-class technique, e.g., using three phases for routine abdomen. These findings may guide clinicians in protocol design, by providing best-practice protocols from a large registry, as well as a method for analyzing data within their own health system. Key Points <Emphasis Type="BoldItalic">Question</Emphasis> Can we identify best-practice imaging protocols from a large CT registry, to offer clinicians a blueprint for optimizing radiation dose in routine abdomen CT? <Emphasis Type="BoldItalic">Findings</Emphasis> Protocols were grouped into clusters using average acquisition parameters. Across clusters, patient size-adjusted radiation dose varied 486–1382 mGy-cm. Lowest-dosed protocols had one phase and low kVp. <Emphasis Type="BoldItalic">Clinical relevance</Emphasis> We used crowd-sourcing and k-means clustering to identify routine abdomen CT protocols that achieve lowest radiation dose, offering a playbook for protocol design from real clinical practice. Non-indicated phases were the greatest driver of excessive radiation dose. Graphical Abstract

Crowd-sourcing optimized abdomen CT protocols from 908,000 examinations in a large radiation dose registry

期刊论文
出版模式：

开放获取
发表日期：

2025年11月24日

查看全文

Rebecca Smith-Bindman,
Taewoon Kang,
Carly Stewart,
Philip W. Chu,
Yifei Wang,
Timothy P. Szczykutowicz

摘要

Objectives

Identify routine diagnostic abdomen CT protocols in adults that achieve lower radiation doses, crowd-sourced from a large CT dose registry.

Materials and methods

We retrospectively captured acquisition parameters, patient diameter, and patient size-adjusted and unadjusted radiation dose (CTDIvol and DLP). A protocol was defined as a unique combination of facility, scanner make/model, CT indication, and protocol name. We used k-means clustering to classify protocols into clusters based on similarity of average acquisition parameters (mAs, pitch, kV, collimation, scan length, phase count), each representing a distinct pattern of protocol design choices. For each cluster, we summarized the mean technical parameters and dose metrics of its constituent protocols.

Results

Analyses included 907,992 exams from 1767 protocols at 132 facilities, grouped in 9 clusters. The number of protocols and exams within each cluster ranged from 62 to 508 and 15,317 to 381,457, respectively. Mean size-adjusted DLP varied threefold across clusters, ranging from 486 to 1382 mGy-cm, with no difference in patient diameter. Lowest-dosed clusters 1 and 2 minimized radiation primarily through low kV (around 100). Cluster 8 had the highest acquisition techniques (mean mAs = 338 and kV = 125), and cluster 9 had the highest dose due to phase (mean = 3.3, versus 1.1–1.6 for the remaining clusters).

Conclusion

The clustering approach offers a new framework for identifying optimized protocols, while highlighting variation in practice and worst-in-class technique, e.g., using three phases for routine abdomen. These findings may guide clinicians in protocol design, by providing best-practice protocols from a large registry, as well as a method for analyzing data within their own health system.

Key Points

Question Can we identify best-practice imaging protocols from a large CT registry, to offer clinicians a blueprint for optimizing radiation dose in routine abdomen CT?

Findings Protocols were grouped into clusters using average acquisition parameters. Across clusters, patient size-adjusted radiation dose varied 486–1382 mGy-cm. Lowest-dosed protocols had one phase and low kVp.

Clinical relevance We used crowd-sourcing and k-means clustering to identify routine abdomen CT protocols that achieve lowest radiation dose, offering a playbook for protocol design from real clinical practice. Non-indicated phases were the greatest driver of excessive radiation dose.

Graphical Abstract