Truace
Subscribe

Multimodal Carotid Ultrasound-based Artificial Intelligence Model for Predicting Acute Ischemic Stroke Risk in Patients with Type 2 Diabetes Mellitus

Rationale and objectives Patients with type 2 diabetes mellitus (T2DM) are at high risk of acute ischemic stroke (AIS), yet accurate risk stratification remains challenging. This study aimed to develop and validate a multimodal model based on carotid ultrasound for improved AIS risk prediction. Materials and methods In this multicenter retrospective study, a total of 480 patients with T2DM who underwent carotid ultrasound were recruited from two centers, with one center (n=394) used for model development and the…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1651· cite & verify
Multimodal Carotid Ultrasound-based Artificial Intelligence Model for Predicting Acute Ischemic Stroke Risk in Patients with Type 2 Diabetes Mellitus
"CT scan Rasmussen's encephalitis" by Mikhail Kalinin is licensed under CC BY-SA 3.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/3.0/.

In brief

Researchers built a model that analyzes carotid ultrasound images of patients with type 2 diabetes. A Swin Transformer framework looked at both plaque detail and surrounding vessel context, and its outputs were merged with radiomic and clinical data to estimate the risk of acute ischemic stroke.

In testing, the combined model outperformed models using a single data type, and readers interpreting scans with AI help scored better. This matters because stroke risk is high in this group yet hard to stratify. The evidence comes from retrospective data at two centers, so its value in routine care is not yet shown.

Main points

  1. The retrospective study included 480 patients with type 2 diabetes who had carotid ultrasound, drawn from two centers.
  2. One center (n=394) supplied model development data, while the other (n=86) served as an independent external test set.
  3. The multimodal model outperformed single-modality models, with AUCs of 0.952 internally and 0.939 externally.
  4. AI assistance raised AUCs for senior readers (0.760 to 0.857 and 0.868 to 0.929) and for junior readers (0.450 to 0.700 and 0.548 to 0.659).
  5. Generalized estimating equation analyses showed significant gains in diagnostic correctness for senior (P=0.00235) and junior (P=0.000343) readers.

The gain

Combining deep learning, radiomic and clinical features from carotid ultrasound predicted acute ischemic stroke in type 2 diabetes patients with AUCs near 0.94-0.95, and AI assistance improved senior and junior readers' diagnostic performance.

The rundown

The design combined three feature types: deep learning features from a dual-scale Swin Transformer capturing local plaque and vascular context, radiomic features, and clinical features. Performance was judged by AUC, calibration, and decision curve analysis.

A reader study compared interpretation with and without AI help. Both senior and junior readers improved, and junior readers started from lower baseline AUCs, with one reported at 0.450 before AI assistance.

The authors conclude that the framework enables improved stroke risk stratification in type 2 diabetes. The external test set was small, as shown by the wide confidence interval of 0.838-0.996.

Sources

  1. Peer-reviewedAcademic Radiology2026-10-09

The debate