What is the clinical evidence from this study?

Study design: Cross-Sectional. Population: High atherosclerotic cardiovascular disease (ASCVD) risk (n=169). Intervention: AI-enhanced point-of-care ultrasound (POCUS) vs. Expert ultrasonologist review. Primary outcome: Per-patient carotid plaque detection agreement (Kappa 0.78, 95% CI 0.69-0.88).

What does this research mean for the field?

After systematic training, general practitioners can accurately detect carotid plaques in high-risk patients using AI-enhanced point-of-care ultrasound, achieving high sensitivity (0.87) and specificity (0.91). Novelty: ClaimNovelty.CONFIRMATORY. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To explore the diagnostic performance of general practitioners (GPs) using AI-enhanced point-of-care ultrasound for carotid plaque detection after systematic training.

May 29, 2026Open Access

Diagnostic Performance of General Practitioners in Carotid Plaque Detection Using AI-Enhanced Point-of-Care Ultrasound After Systematic Training

Key Result

General practitioners using AI-enhanced point-of-care ultrasound achieved high diagnostic agreement (kappa 0.78) and sensitivity (0.87) for detecting carotid plaque compared to expert ultrasonologists.

Key Points

To explore the diagnostic performance of general practitioners (GPs) using AI-enhanced point-of-care ultrasound for carotid plaque detection after systematic training.
Conducted in Shanghai with 7 GPs who received systematic training.
169 high-risk patients underwent free carotid plaque screening during outpatient visits.
Evaluated diagnostic performance using sensitivity, specificity, PPV, NPV, and κ statistic for agreement.
At the per-patient level, sensitivity was 0.87 (95% CI, 0.80-0.95) and specificity was 0.91 (95% CI, 0.86-0.97).
PPV was 0.89 (95% CI, 0.82-0.96) and NPV was 0.89 (95% CI, 0.83-0.96).
κ value was 0.78 (95% CI, 0.69-0.88), indicating high agreement with the gold standard.

Study Design

Type

Cross-Sectional (n=169)

Blinding

Single-blind

Multicenter

Yes

Structured PICO

Does AI-enhanced POCUS performed by systematically trained general practitioners accurately detect carotid plaque in patients with high ASCVD risk?

Population

169 patients with high atherosclerotic cardiovascular disease risk, mean age 69.6 years, 37.9% male, recruited during outpatient visits in Shanghai.

Intervention

Carotid plaque screening using AI-enhanced point-of-care ultrasound (POCUS) performed by systematically trained general practitioners.

Comparator

Gold standard diagnostic assessment

Outcome

Diagnostic performance (sensitivity, specificity, PPV, NPV, and κ statistic) for carotid plaque detection at per-patient and per-vessel levels.surrogate

Systematically trained general practitioners can effectively and accurately perform carotid plaque screening using AI-enhanced POCUS in high-risk elderly populations.

Main Result

Effect estimate: Kappa 0.78 (95% CI 0.69-0.88)

Limitations

Risk of incorporation bias by using ultrasound as the reference standard instead of MRI or CTA
Lack of an unassisted control arm limits the ability to disentangle the incremental value of AI from GPs' scanning proficiency
Small sample size of only 7 self-selected, relatively young and experienced GPs
Homogeneous patient population restricted to high-risk elderly individuals in Shanghai
Exclusion of vertebral and innominate arteries from the screening protocol

Abstract

PURPOSE While point-of-care ultrasound (POCUS) has been integrated into daily practice by general practitioners (GPs) in some countries, there is a paucity of literature documenting its use by Chinese GPs. Additionally, artificial intelligence (AI)-assisted diagnosis is increasingly applied to imaging equipment. This study aims to explore the diagnostic performance of GPs in carotid plaque detection using AI-enhanced POCUS after systematic training. METHODS This diagnostic accuracy trial was conducted in Shanghai, where 7 GPs received systematic training then recruited patients with high atherosclerotic cardiovascular disease risk during outpatient visits for free carotid plaque screening. We comprehensively evaluated their diagnostic performance using 2 approaches: per-patient (ie, assessed whether patients had any carotid plaque) and per-vessel (ie, determined plaque presence in any carotid vessel). Evaluation metrics included sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the κ statistic for agreement. RESULTS A total of 169 patients (mean age 69.6 years SD 5.76; 37.9% male) were included in this study. At the per-patient level, sensitivity was 0.87 (95% CI, 0.80-0.95); specificity was 0.91 (95% CI, 0.86-0.97); PPV was 0.89 (95% CI, 0.82-0.96); NPV was 0.89 (95% CI, 0.83-0.96); and the κ value was 0.78 (95% CI, 0.69-0.88), indicating high agreement with the gold standard. The diagnostic performance evaluated at the per-vessel level was comparable to that at the per-patient level. CONCLUSION Our findings indicate that, after systematic training, GPs can effectively perform carotid plaque screening using AI-enhanced POCUS within a clinical workflow in high-risk elderly populations.

Bookmark

View Full Paper