The role of the combined [18F]FDG PET/CT carotid artery and brain studies in stroke prediction: an AI-based prediction of stroke after a 10-year observation period

Q: What is the clinical evidence from this study?

Study design: Cohort. Population: Symptomatic carotid artery stenosis (n=43). Intervention: AI-based Bayesian network integrating [18F]FDG PET/CT carotid and brain metabolic activity with clinical and ultrasound data. Primary outcome: Occurrence of ischemic or hemorrhagic stroke over 10-year follow-up (AUC 0.76, 95% CI 0.57-0.92, p=0.03).

Q: What does this research mean for the field?

An AI-based Bayesian network integrating carotid artery and brain [18F]FDG PET/CT with ultrasound data predicts 10-year stroke risk with 77% accuracy and an AUC of 0.76. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Beata Ewa Chrapko; Jakub Mitura; Agata Danilczuk; Marek Chrapko; Piotr Terlecki; Sebastian Piotr Wiśniewski

doi:10.5603/nmr.111509

April 20, 2026Nuclear Medicine ReviewOpen Access

The role of the combined [18F]FDG PET/CT carotid artery and brain studies in stroke prediction: an AI-based prediction of stroke after a 10-year observation period

View Full Paper

Ask AI

Bookmark

Share

Key result

AI model integrating carotid-brain PET/CT with ultrasound predicts 10-year stroke risk with ~77% accuracy.

AUC 0.76
95% CI 0.57-0.92
P=0.03
n=43

Why the study?

This study evaluated whether integrating metabolic biomarkers from [18F]FDG PET/CT scans of carotid plaques and the brain using an AI-based model can assess long-term (10-year) stroke risk.

Does an AI-based Bayesian network integrating [18F]FDG PET/CT metabolic biomarkers of carotid plaques and the brain with clinical and ultrasound data predict long-term (10-year) stroke occurrence in patients with symptomatic carotid artery stenosis (>50%)?

Key Points

To evaluate whether an artificial intelligence model integrating metabolic biomarkers from [18F]FDG PET/CT imaging of carotid plaques and the brain can assess 10-year stroke risk.
Enrolled 43 symptomatic patients (median age 66 years) with carotid plaques in a prospective study who underwent carotid Doppler ultrasound and brain [18F]FDG PET/CT imaging.
Developed a Bayesian network integrating maximum standardized uptake values (SUVmax) and clinical variables to forecast stroke events over a 10-year observation period.
Evaluated model performance using 5-fold cross-validation.
Over the 10-year follow-up, 21 patients experienced a stroke, with a significant association identified between increased SUVmax in the right common carotid artery and subsequent stroke (p = 0.03).
The Bayesian network achieved an overall accuracy of 77% (24/31 cases; 95% CI: 0.59–0.90) and an area under the curve of 0.76 (95% CI: 0.57–0.92).
The model correctly classified 83% of individuals without a stroke and 73.6% of individuals with a stroke, while sensitivity analysis on the full unedited dataset (n = 35) showed 65.7% accuracy and an AUC of 0.61.

Study Design

Type

Cohort (n=43)

Multicenter

No

Structured PICO

Does an AI-based Bayesian network integrating [18F]FDG PET/CT metabolic biomarkers of carotid plaques and the brain with clinical and ultrasound data predict long-term (10-year) stroke occurrence in patients with symptomatic carotid artery stenosis (>50%)?

P

Population

43 symptomatic adults (median age 66 years, 44.2% female) with >50% carotid artery stenosis referred for surgical evaluation, followed for a median of 10 years.

E

Exposure

Combined [18F]FDG PET/CT imaging of the brain (1 hour post-injection of 3.5 MBq/kg [18F]FDG) and carotid arteries (3 hours post-injection), integrated with carotid Doppler ultrasonography and clinical data into an AI-based probabilistic Bayesian network.

O

Outcome

Occurrence of ischemic or hemorrhagic stroke over a median 10-year follow-up period (range 8–10 years), assessed via telephone interviews.hard clinical

Main Result

Effect estimate: AUC 0.76 (95% CI 0.57-0.92)

p-value: p=0.03

This proof-of-concept study demonstrates that integrating carotid plaque and brain metabolic activity from [18F]FDG PET/CT into a Bayesian network may help predict 10-year stroke risk in patients with symptomatic carotid stenosis.

Limitations

Very small sample size (43 enrolled, 31 evaluated in final cross-validation)
High risk of model instability and overfitting
Reliance on self-reported outcome data via telephone interviews without systematic clinical and radiological adjudication
Inability to reliably verify whether strokes were ipsilateral to the evaluated carotid plaque
No correction for multiple hypothesis testing
Exclusion of medication data due to high missingness
Outlier removal performed on the entire dataset prior to cross-validation
Very small dataset (43 enrolled, 31 evaluated in final cross-validation)
No corrections were made for multiple hypothesis testing, making findings strictly hypothesis-generating
Follow-up relied mainly on self-reported telephone interviews with patients or families without systematic clinical and radiological adjudication
High missingness of medication data (ASA, statins, anticoagulants) requiring exclusion from modeling
LOF preprocessing and outlier removal were performed on the entire dataset prior to cross-validation due to the small sample size

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

May support AI-PET integration for stroke risk stratification in symptomatic carotid disease; leaves open prospective validation in larger cohorts.

Abstract

BACKGROUND: Long-term (10-year) stroke risk may be assessed by integrating metabolic biomarkers from [¹⁸F]FDG PET/CT scans of carotid plaques and the brain using an artificial intelligence (AI)-based model. MATERIAL AND METHODS: A prospective study enrolled 43 symptomatic patients (median age 66 years) with carotid plaques. Participants underwent Doppler ultrasound of carotid arteries and [¹⁸F]FDG PET/CT of the brain. A Bayesian network was developed to predict stroke occurrence, incorporating imaging (SUVmax) and clinical data. Model performance was evaluated using 5-fold cross-validation and a 10-year follow-up. RESULTS: During follow-up, 21 patients experienced a stroke. A statistically significant association was observed between increased metabolic activity (SUVmax) in the right common carotid artery and subsequent stroke (p = 0.03). The AI model achieved an overall accuracy of 77% and an area under the curve (AUC) of 0.76. CONCLUSIONS: This proof-of-concept study suggests that AI-integrated multimodal data, specifically [¹⁸F]FDG PET/CT imaging of carotid and brain metabolism, may provide useful exploratory biomarkers for long-term stroke risk assessment.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Chrapko et al. (2026) conducted a cohort in Symptomatic carotid artery stenosis (n=43). AI-based Bayesian network integrating [18F]FDG PET/CT carotid and brain metabolic activity with clinical and ultrasound data was evaluated on Occurrence of ischemic or hemorrhagic stroke over 10-year follow-up (AUC 0.76, 95% CI 0.57-0.92, p=0.03). An AI-based Bayesian network integrating carotid and brain [18F]FDG PET/CT with ultrasound data predicted 10-year stroke risk with 77% accuracy and an AUC of 0.76 (95% CI 0.57-0.92).

synapsesocial.com/papers/6af0e999de503f3774140a39 https://doi.org/10.5603/nmr.111509

View Full Paper

Ask AI

Bookmark

Share

The role of the combined [18F]FDG PET/CT carotid artery and brain studies in stroke prediction: an AI-based prediction of stroke after a 10-year observation period

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

The role of the combined [18F]FDG PET/CT carotid artery and brain studies in stroke prediction: an AI-based prediction of stroke after a 10-year observation period

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading