Researchers have developed and validated an advanced AI-driven ultrasound framework that assesses carotid plaque vulnerability and predicts cerebrovascular risk from routine B-mode scans. Evaluated across 6,618 participants, the tool successfully flags high-risk patients, offering a vital new approach to stroke prevention and cardiovascular disease management.
Multicenter AI Framework Evaluates Carotid Plaque Vulnerability
Routine carotid B-mode ultrasound is widely used for plaque assessment, but vulnerability evaluation remains subjective, and its prognostic value is uncertain. To address this clinical gap, researchers developed and validated a cascaded multitask AI framework designed for plaque detection, characterization of five B-mode features, and vulnerability assessment.
The multicenter cohort included 6,618 participants at high cardiovascular risk and 38,090 images from four centers. Plaque presence and B-mode features were evaluated against expert B-mode consensus, whereas vulnerability was evaluated against multimodal expert consensus. Internal and external AUCs were 0.95 and 0.96 for plaque detection, 0.85–0.94 and 0.81–0.92 for feature characterization, and 0.90 and 0.88 for vulnerability classification. The model outperformed six independent B-mode readers and showed greater net benefit in decision-curve analysis.
Prognostic validation included 2,174 participants, during which 277 ischemic cerebrovascular events occurred over a median of 37 months. AI-defined high-risk status remained independently associated with events after adjustment for clinical risk factors (hazard ratio, 2.09; 95% confidence interval, 1.69–2.58), and adding the AI score improved the C-index from 0.769 to 0.795 in the published findings.
Testing AI-Assisted Ultrasound in Primary Care Clinics
Parallel research published in the Annals of Family Medicine evaluated whether general practitioners could use AI-assisted portable ultrasound to detect carotid artery plaque in community health centers. Conducted in Shanghai, China, the feasibility study involved seven general practitioners who completed a training program combining classroom instruction with hands-on practice using an AI-enhanced point-of-care ultrasound, known as POCUS.
The practitioners recruited 169 patients at high risk for atherosclerotic cardiovascular disease during regular outpatient visits, offering them free carotid plaque screening. Each patient was scanned by both a general practitioner and a senior ultrasound specialist, with the latter group independently reviewing all stored recordings and images to establish a benchmark.
Looking at whether patients did or did not have plaque overall, the general practitioners correctly identified approximately 87% of patients who had plaque and correctly ruled it out in approximately 91% of patients who did not. Their findings showed a high level of agreement with the senior ultrasound specialist benchmark. However, diagnostic performance was not uniform. Among vessels with confirmed plaque, missed cases were concentrated at the fork where the carotid artery splits (18.6%, 22/118), which authors attribute to potential limitations of the AI system in that area or gaps in how practitioners incorporated the AI feedback into their workflow. Misidentifying plaque in patients who did not actually have it was less common. Each scan took roughly 8 minutes to complete, highlighting potential scheduling hurdles in fast-paced primary care settings where consultation times tend to be short.
“The widespread implementation of this AI-assisted POCUS approach will necessitate standardized, operator-specific training modules, as well as further validation in larger, more heterogeneous general practitioner cohorts, to ensure uniformly reliable performance.”
Study authors, via Annals of Family Medicine
Precision Medicine and the TAXINOMISIS Risk Stratification Tool
To combat the disease variability that impedes effective clinical decision-making, the EU-funded TAXINOMISIS project developed a novel model that categorizes patients based on their risk of developing carotid artery disease and experiencing cerebrovascular events manifested as stroke, transient ischaemic attack, or transient loss of vision.
The tool encompasses AI-based components that integrate clinical information, personalized patient data, plaque and brain images, blood flow patterns, and novel biomarkers. Computer models and simulations predict silent brain lesions, cardiovascular events, plaque vulnerability, and plaque evolution and rupture. The risk stratification tool assesses the risk of carotid artery disease by using non-imaging data as input to evaluate the likelihood of high-risk plaques, while also linking the condition of the carotid artery to the presence of brain lesions and predicting the risk of new brain lesions in patients.
Furthermore, the stratification tool includes a blood flow component applied to predict the risk for plaque progression and estimate the risk of plaque rupture. Medical professionals can utilize either a complex model taking into account structural and compositional factors of the arterial wall and plaque, or a simplified model focusing only on the lumen, allowing them to choose the appropriate level of analysis based on available data, resources, and the clinical context. Advanced AI techniques are also used to reconstruct the 3D geometry of arteries and to characterize plaque through magnetic resonance and ultrasound imaging data.
“TAXINOMISIS has designed a rational new approach to address key clinical gaps in carotid artery disease in terms of predicting progression, need for follow-up and selection of appropriate interventions.”
Dimitrios Fotiadis, project coordinator
Integrating Advanced AI Models Into Daily Practice
The development of these AI frameworks highlights a broader push to move beyond the limitations of standard screening guidelines. By speeding up the diagnosis process and reducing the need for more intrusive tests or treatments, the TAXINOMISIS tool aims to contribute to lower healthcare expenditures while increasing health system efficiency.
Researchers plan to further improve platform algorithms and models using input from physicians, users, and data from clinical trials and real-world cases. Joint ventures and collaborations with other organizations will be instrumental in integrating the stratification tool into healthcare systems or combining it with other available tools, paving the way for enhanced clinical decision-making through accurate risk stratification.
