Carotid ultrasound AI study reports promising internal results, with validation still needed
Published in Artificial Intelligence, Journals, Scanners and Imaging.
A study published in BMC Medical Imaging on September 26, 2026, evaluated machine learning combining carotid plaque ultrasound measurements with clinical factors. Researchers prospectively enrolled 231 inpatients between December 2022 and December 2024.
Patients were grouped by ischemic stroke status using recent brain imaging. The dataset was randomly divided into 162 training patients and 69 test patients. A random forest model performed best in the internal test set, with an area under the ROC curve of 0.866, accuracy of 81%, sensitivity of 80% and specificity of 83%.
The five selected features were plaque thickness, near-shoulder shear-wave elastography, plaque neovascularization, triglycerides and smoking history. These findings suggest that combining ultrasound and clinical information may help characterize stroke-associated plaque patterns.
However, the test patients came from the same enrolled dataset. The reported results do not establish performance in an independent clinical population or prediction of future strokes. External and prospective outcome validation would be needed before treating this model as a clinical risk forecasting tool.
Source: Ni and colleagues, BMC Medical Imaging, September 26, 2026.