Computed tomographic angiography-derived radiomics features of the plaque and lumen region had better predictive value for stroke recurrence than radiomics features of the plaque region alone in patients with symptomatic intracranial atherosclerotic stenosis.
Key Findings
Results
Radiomics models based on the combined plaque and lumen region outperformed plaque-region-only models for predicting stroke recurrence in both training and testing sets.
Plaque and lumen region models achieved area under the curve (AUC) of 0.86–0.87 in the training set versus 0.79–0.80 for plaque-region-only models.
In the testing set, plaque and lumen region models achieved AUC of 0.74–0.81 versus 0.58–0.62 for plaque-region-only models.
The superior performance of plaque and lumen region models was consistent with internal validation results using repeated cross-validation.
A similar trend was observed across performance metrics, calibration analysis, and reclassification indices.
Results
No significant differences in baseline clinical variables were observed between patients who experienced the primary outcome and those who did not.
The primary outcome was same-territory ischemic stroke recurrence within 1 year.
Clinical variables were collected at baseline for all 203 patients.
This finding suggests that clinical variables alone may be insufficient to discriminate stroke recurrence risk in this population.
All patients received guideline-recommended medical treatment.
Results
A small subset of radiomics features was selected from a large pool of extracted features for model development.
A total of 1046 radiomics features were extracted from computed tomographic angiography source images.
4 radiomics features from the plaque region alone were selected for model development.
5 radiomics features from the combined plaque and lumen region were selected for model development.
Four machine learning algorithms were used to develop and test predictive models.
Methods
The study enrolled patients with symptomatic intracranial atherosclerotic stenosis of 50%–99% confirmed by computed tomographic angiography from two centers.
The total cohort comprised 203 patients with a median age of 60 years.
Patients had experienced ischemic stroke or transient ischemic attack with symptomatic intracranial atherosclerotic stenosis.
Patients were randomly split into a training set (n=121, 60%) and a testing set (n=82, 40%).
This was a substudy of a cohort study from 2 centers.
Conclusions
Computed tomographic angiography-derived radiomics features of the plaque and lumen region offer a promising tool for stroke risk stratification in symptomatic intracranial atherosclerotic stenosis.
The combined plaque and lumen region radiomics models demonstrated better discrimination, calibration, and reclassification than plaque-only models.
The predictive advantage was consistent across all four machine learning algorithms tested.
The findings support the use of radiomics as an adjunct to clinical assessment for identifying high-risk patients.
The primary outcome assessed was same-territory ischemic stroke recurrence within 1 year.
What This Means
This research examined whether detailed imaging analysis of arterial plaques — using a technique called radiomics — could predict which stroke patients are most likely to have another stroke within a year. The study focused on patients who had narrowing of brain arteries (intracranial atherosclerotic stenosis) and had already experienced a stroke or mini-stroke. Using CT angiography scans, the researchers extracted over 1,000 mathematical features describing the texture, shape, and intensity of the diseased artery segments, then used machine learning to identify which features best predicted stroke recurrence.
A key finding was that analyzing both the plaque (the buildup in the artery wall) and the lumen (the open channel where blood flows) together produced much better predictions than analyzing the plaque alone. Models combining both regions achieved accuracy scores (AUC) of 0.74–0.81 on new patients, compared to only 0.58–0.62 for plaque-only models. Notably, standard clinical information — such as patient age and medical history — did not significantly differ between patients who had a recurrent stroke and those who did not, suggesting that these traditional measures alone are not enough to identify who is at highest risk.
This research suggests that radiomics analysis of CT angiography scans, particularly when the full arterial cross-section including the blood channel is analyzed, could become a useful tool to help doctors identify stroke patients who are at higher risk of having another stroke. This could potentially allow for more personalized treatment strategies. Because the study was relatively small (203 patients) and conducted at only two centers, further validation in larger and more diverse populations will be needed before this approach could be used in routine clinical care.
Li Z, Abrigo J, Tian X, Li S, Liu Y, Liu Y, et al.. (2026). Computed Tomographic Angiography-Based Radiomics Predicted Stroke Recurrence in Patients With Symptomatic Intracranial Atherosclerotic Stenosis.. Journal of the American Heart Association. https://doi.org/10.1161/JAHA.125.048337