Monday, December 1 | 1:40 p.m.-1:50 p.m. | SSVA02-2 | Room S503
This vascular imaging session presents an AI-facilitated management strategy for early-stage coronary atherosclerosis (CAS) in asymptomatic populations.
The strategy incorporates a machine-learning (ML) model developed from traditional cardiovascular risk factors and quantitative coronary CT angiography (CCTA)-derived plaque features, according to a team led by Jiansong Ji, MD, from the Fifth Affiliated Hospital of Wenzhou Medical College in Zhejiang, China.
In advance of RSNA 2025, Ji and colleagues reported that the model produces quantifiable insights and predicts short-term plaque progression. It may enable personalized risk stratification before CAS onset, they added.
A total of 329 adults with early-stage CAS were included in the retrospective cohort study, of whom 57% were male with a mean age of 63 years. For their study, Ji and colleagues analyzed the cohort's serial CCTA data using the model.
During the median follow-up period of 56 months, the cohort demonstrated distinct plaque progression patterns in early-stage disease, the group reported, noting that the findings could influence treatment decisions that might delay atherosclerotic progression.
For the session, presenter Yanping Su will explain how the group used SHapley Additive exPlanation (SHAP) analysis, a feature-based interpretability method. Among the eight ML algorithms, random forest performed best, according to the results.
The group's preliminary evaluation metrics included a training area under the curve (AUC) of 0.94, compared with a testing AUC of 0.84.
The model outperformed traditional methods, according to the group. SHAP analysis identified key progression drivers: fibrofatty plaque volume, plaque composition heterogeneity, and monocyte count, they said.
Drop in after lunch to learn more.
