Carotid duplex ultrasound radiomics significantly enhances the ability to predict adverse outcomes in acute ischemic stroke, offering a precision-based approach that supports personalized management strategies.
Key Findings
Results
A radiomic model developed from carotid duplex ultrasound images demonstrated high accuracy and clinical utility for predicting adverse outcomes 3 months post-stroke in AIS patients.
The study enrolled 105 AIS patients at the Stroke Center of the First Affiliated Hospital of Soochow University.
CDU images were processed using artificial intelligence to extract 1,477 radiomic features.
Key features were selected from the 1,477 extracted features to develop the predictive model.
Clinical utility was confirmed by decision curve analysis.
Outcomes were assessed at 3 months post-stroke.
Results
Inter-observer reliability for assessing plaque vulnerability using CDU radiomics was excellent.
Cohen's kappa coefficient was 0.93, indicating consistent assessments of plaque vulnerability across different observers.
A kappa of 0.93 is considered excellent agreement in standard interpretive frameworks.
This reliability measure supports the reproducibility of the radiomic approach in clinical settings.
Methods
Carotid duplex ultrasound radiomics was applied to characterize carotid plaque vulnerability in AIS patients.
CDU images were obtained from 105 AIS patients enrolled at a single stroke center.
Artificial intelligence was used to process CDU images and extract radiomic features.
A total of 1,477 radiomic features were extracted from the CDU data.
The radiomic approach was used to assess plaque vulnerability as a component of outcome prediction.
Conclusions
The CDU radiomics-based predictive model supports personalized management strategies by allowing targeted interventions based on specific radiomic profiles.
The model predicts adverse outcomes at 3 months post-stroke.
The authors indicate the approach could improve clinical outcomes through targeted interventions.
The study frames the radiomic model as enabling a 'precision-based approach' to AIS management.
The model was characterized as having high accuracy and clinical utility confirmed by decision curve analysis.
What This Means
This research suggests that analyzing ultrasound images of the carotid arteries (the major blood vessels in the neck) using advanced computer-based techniques can help predict how well stroke patients will recover. The researchers took ultrasound images from 105 patients who had experienced an acute ischemic stroke and used artificial intelligence to extract nearly 1,500 detailed measurements—called radiomic features—from those images. They then used those measurements to build a model that predicted patient outcomes three months after the stroke. The model performed well, and two different observers analyzing the same images came to very consistent conclusions, with a high agreement score (kappa of 0.93).
This research matters because predicting stroke outcomes early can help doctors make better decisions about treatment. Currently, clinicians rely on a combination of clinical assessments and standard imaging, but identifying who is at highest risk for poor outcomes remains challenging. This study suggests that detailed computer analysis of routine carotid ultrasound images—a widely available and non-invasive imaging tool—could add meaningful predictive information beyond what is already used in clinical practice.
If these findings are validated in larger, independent groups of patients, carotid ultrasound radiomics could eventually be incorporated into clinical workflows to help tailor treatment plans to individual patients based on their specific imaging profiles. The authors describe this as a step toward more personalized stroke management, though further research would be needed before such an approach could be adopted in routine care.
Zhang L, Cao H, Miao L, Zhang X, Shi G. (2026). Advanced radiomics techniques applied on carotid duplex ultrasound data predict adverse outcomes in acute ischemic stroke.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1878428