The combined machine learning model achieved an AUC of 0.86 for detecting symptomatic carotid atherosclerosis, outperforming the clinical model (AUC 0.67, p=0.03).
Does a combined machine learning model integrating CTA radiomics of carotid plaques and perivascular adipose tissue with clinical data improve the detection of symptomatic carotid atherosclerosis compared to a clinical model alone?
229 patients with extracranial carotid atherosclerotic plaques who underwent computed tomography angiography (CTA), mean age 68.80, 86.46% male. Patients were divided into symptomatic (n=127) and asymptomatic (n=102) groups based on the occurrence of cerebrovascular events within two weeks prior to the CTA examination.
Combined machine learning model (Random Forest) integrating clinical data, perivascular adipose tissue (PVAT) radiomics, and plaque radiomics extracted from CTA images
Clinical machine learning model alone (based on traditional clinical features)
Detection of symptomatic patients (defined by the occurrence of cerebrovascular events within two weeks prior to the CTA examination) measured by Area Under the Curve (AUC)
Integrating radiomics of carotid plaques and perivascular adipose tissue with clinical data using a Random Forest machine learning model significantly enhances the diagnostic detection of symptomatic carotid atherosclerosis.
Absolute Event Rate: 0% vs 0%
Abstract Background This study aimed to develop a machine learning (ML) model based on radiomics features of carotid plaques and perivascular adipose tissue (PVAT) on computed tomography angiography (CTA) to detect symptomatic carotid atherosclerosis. Methods This retrospective study included patients with extracranial carotid atherosclerotic plaques who underwent CTA between January 2022 and January 2024. Patients were divided into symptomatic and asymptomatic groups based on the occurrence of cerebrovascular events within two weeks prior to the CTA examination. Five ML models were constructed to identify symptomatic patients: clinical, PVAT radiomics, plaque radiomics, PVAT and plaque radiomics, and combined model. The most robust model was selected for Shapley Additive Explanations (SHAP) analysis to visualize the prediction process. Results The study cohort consisted of 229 patients (127 symptomatic; 102 asymptomatic). The Random Forest models demonstrated the best performance in detecting symptomatic patients. In the test cohort, the area under the curve (AUC) of the combined model (0.86; 95% confidence interval CI: 0.74–0.95) was significantly higher than that of the clinical model (AUC: 0.67, 95% CI: 0.50–0.81; p = 0.03), but similar to that of the PVAT and plaque radiomics model (AUC: 0.82, 95% CI: 0.70–0.93; p = 0.65). SHAP analysis of the combined model identified carotid plaque texture features and cholesterol levels as key factors in detecting symptomatic patients. Conclusions Integrating radiomics of carotid plaques and PVAT with clinical data enhances the detection of symptomatic patients.
Building similarity graph...
Analyzing shared references across papers
Loading...
Yulu Yang
University of Shanghai for Science and Technology
Jianyong Wei
Shanghai Sixth People's Hospital
Xiaoer Wei
Shanghai Sixth People's Hospital
BMC Medical Imaging
University of Shanghai for Science and Technology
Shanghai Sixth People's Hospital
Beijing Chaoyang Emergency Medical Center
Building similarity graph...
Analyzing shared references across papers
Loading...
Yang et al. (Mon,) reported a other. The combined machine learning model achieved an AUC of 0.86 for detecting symptomatic carotid atherosclerosis, outperforming the clinical model (AUC 0.67, p=0.03).
synapsesocial.com/papers/6963222f91e05aa366cb8c38 — DOI: https://doi.org/10.1186/s12880-025-02113-1