# Reliability of Handheld Ultrasound Assessment of Brachial Artery Flow-Mediated Dilation Using AI-Assisted Automated Analysis in Postmenopausal Women

## TL;DR

This study shows that an AI-assisted handheld ultrasound system reliably measures blood vessel dilation in postmenopausal women, offering a promising tool for cardiovascular research.

## Contribution

The novel contribution is demonstrating the reliability of an AI-assisted workflow using a YOLOv12 model for automated FMD assessment in postmenopausal women.

## Key findings

- Baseline and peak diameters showed good between-day reliability with ICCs of 0.81 and 0.76.
- FMD% had high ICC (0.87) but higher variability (19.09% CV), indicating individual differences in vascular responses.
- Bland–Altman analysis confirmed minimal bias in FMD% measurements between test days.

## Abstract

Background and Objectives: Endothelial dysfunction is an early indicator of cardiovascular disease and is commonly assessed using flow-mediated dilation (FMD). Although handheld ultrasound (HHUS) devices improve measurement accessibility, image analysis for conventional flow-mediated dilation (FMD) assessment remains time-consuming and highly operator-dependent. This study aimed to evaluate the between-day test–retest reliability of an AI-assisted brachial artery image analysis workflow integrating HHUS imaging with a YOLOv12 deep learning model in postmenopausal women. Materials and Methods: Seventeen postmenopausal women aged 55–70 years completed two flow-mediated dilation assessments conducted seven days apart. Brachial artery images were acquired using a standardized FMD protocol with a handheld ultrasound system. An AI-assisted image analysis workflow based on a YOLOv12 deep learning model was used to automatically measure baseline diameter (Dbase), peak diameter (Dpeak), absolute FMD (FMDabs), and relative FMD (FMD%). Between-day reliability was evaluated using intraclass correlation coefficients (ICCs), coefficients of variation (CVs), and Bland–Altman analysis. Results: Good between-day repeatability was observed for baseline and peak diameters, with ICCs of 0.81 and 0.76 and low CVs (3.26% and 3.22%), respectively. Functional vascular outcomes also demonstrated good reliability, with ICCs of 0.81 for FMDabs and 0.87 for FMD%. However, higher CVs were observed for FMDabs (17.15%) and FMD% (19.09%), indicating substantial inter-individual variability. Bland–Altman analysis showed a small mean difference for FMD% (0.34%), with no evidence of systematic bias. Conclusions: An AI-assisted HHUS image analysis workflow integrating a YOLOv12 deep learning model demonstrates acceptable between-day reliability for diameter-based and dilation-based measures of flow-mediated dilation in postmenopausal women. While variability in functional responses exists, the proposed system is feasible for research-oriented vascular assessment, providing a methodological foundation for future validation and clinical translation studies.