AI-enhanced handheld ultrasound using a super-resolution reconstruction model (Hyper-CycleGAN) improved carotid plaque detection to 94.8% and stenosis grading agreement (weighted κ 0.836) compared to standard handheld ultrasound in community screening settings.
Key Findings
Results
AI-enhanced handheld ultrasound detected 94.8% of carotid plaques in community screening, compared to 87.6% using standard handheld ultrasound.
The community screening cohort included 117 participants with 153 plaques.
A difference of 7.2 percentage points in plaque detection rate was observed between AI-enhanced and standard HHUS.
Portable ultrasound served as the reference standard in the community setting.
The AI model applied was the Hyper-CycleGAN super-resolution reconstruction model.
Results
Stenosis grading agreement with the reference standard substantially improved with AI enhancement, with weighted κ increasing from 0.492 to 0.836.
Quadratic-weighted Cohen's κ was used to assess agreement in stenosis grading.
Standard HHUS had a weighted κ of 0.492, indicating moderate agreement.
AI-enhanced HHUS achieved a weighted κ of 0.836, indicating strong agreement.
Bland-Altman analysis and intraclass correlation coefficient were also used to evaluate agreement with the reference standard.
Results
Sensitivity for identifying ultrasound-defined vulnerable plaques increased from 47.4% to 63.2% with AI enhancement, while specificity remained high.
Baseline sensitivity for vulnerable plaque detection using standard HHUS was 47.4%.
AI-enhanced HHUS increased vulnerable plaque sensitivity to 63.2%, an absolute increase of 15.8 percentage points.
Specificity was reported to remain high with AI enhancement, though exact values are not provided in the abstract.
Vulnerable plaque identification is clinically important for stroke risk stratification.
Methods
The Hyper-CycleGAN super-resolution reconstruction model was validated in a clinical cohort prior to community deployment.
The validation cohort consisted of 127 patients with 198 carotid plaques.
Validation preceded application in the community screening setting of 117 participants.
The model was developed specifically for enhancing handheld ultrasound image quality.
Diagnostic performance was evaluated using detection rates, quadratic-weighted Cohen's κ, confusion matrices, and standard metrics.
Conclusions
AI-enhanced handheld ultrasound was proposed as a tool to support more accurate risk stratification and referral decisions in primary care and underserved community settings.
The study framed AI-enhanced HHUS as supporting stroke prevention strategies in low-resource or underserved communities.
The approach targets early identification of carotid atherosclerotic plaques, which is described as critical for stroke prevention in primary care.
Ischemic stroke is cited as a leading cause of disability and death, motivating community-level screening.
The authors concluded the approach could expand access to low-cost stroke prevention strategies.
What This Means
This research suggests that adding artificial intelligence to handheld ultrasound devices can significantly improve the ability to detect dangerous plaques in the carotid arteries (the major blood vessels in the neck that supply the brain) during community health screenings. The researchers developed an AI model called Hyper-CycleGAN that enhances the image quality produced by small, portable ultrasound devices, making them more comparable to higher-end equipment. When tested on 117 community participants, the AI-enhanced device correctly identified nearly 95% of plaques, compared to about 88% without AI assistance, and was much better at correctly categorizing how severely the arteries were narrowed.
One particularly important finding was the improvement in detecting 'vulnerable' plaques — the type most likely to break off and cause a stroke. The AI boosted sensitivity for these dangerous plaques from about 47% to 63%, meaning significantly fewer high-risk plaques were missed. Agreement between the handheld device's stenosis grading and the reference standard also improved dramatically, moving from moderate to strong agreement on a standard statistical scale.
This research matters because strokes are a leading cause of death and disability worldwide, and catching artery disease early in community settings — rather than requiring patients to visit specialized hospitals — could help more people receive timely preventive treatment. Handheld ultrasound devices are already relatively affordable and portable, and this research suggests that AI enhancement could make them accurate enough for meaningful stroke risk screening in underserved or rural communities where access to advanced medical imaging is limited.
Lan J, Liu S, Chen H, Zhang T, Zeng P, Li J, et al.. (2026). AI-Enhanced Super-Resolution Handheld Ultrasound for Carotid Plaque Detection in Community Screening.. Annals of family medicine. https://doi.org/10.1370/afm.250838