Wearable Ultrasound Patch Uses AI to Detect Elevated Central Venous Pressure

Central venous pressure (CVP) guides fluid and vasopressor therapy in critically ill patients, yet invasive catheterization carries procedural risks and is not always feasible. Ultrasound assessments of jugular venous congestion can help, but they are intermittent and operator dependent. Clinicians need a comfortable, repeatable method that supports continuous, automated interpretation at the bedside. Researchers have now developed an AI-enabled wearable ultrasound system that estimates CVP noninvasively from the neck and automates image analysis for decision support.

The technology combines a soft, neck‑worn ultrasound patch with dedicated artificial intelligence and was evaluated at Shanghai Sixth People’s Hospital (affiliated with Shanghai Jiao Tong University) and Shanghai Tenth People’s Hospital. The patch images the right internal jugular vein (IJV) and the common carotid artery (CCA), two vessels that reflect right atrial pressure. It is designed for prolonged wear in intensive care units and other settings where invasive monitoring may be contraindicated or delayed.

The 128-element linear-array probe operates at a center frequency of 8.5 MHz to balance imaging resolution and tissue penetration. Dual-layer acoustic matching and tailored backing provide broad bandwidth, while a solid hydrogel interface with silicone encapsulation helps maintain stable acoustic coupling and improve skin comfort. In feasibility testing, image quality remained stable over 24 hours, supporting continuous cine-loop acquisition for subsequent automated analysis.

To process this continuous image stream, the team developed a semi-supervised segmentation model called the dual-decoder spatiotemporal attention network (DSTA-Net). Only about 10% of frames require manual labeling at peak and trough internal jugular vein (IJV) dilation, while the remaining frames are converted into learning signals through a dual-decoder consistency scheme combined with temporal attention.

Across internal and external test sets, DSTA-Net outperformed UNet, Swin-UNet, DeepLabV3+, and leading semi-supervised methods. It achieved Dice scores of 83.5% and 75.8% for IJV segmentation, along with Spearman correlations above 0.88 compared with expert measurements and Bland-Altman percentage errors well below 30%.

The vascular indices automatically extracted from these segmented images, including maximum and minimum IJV area, common carotid artery (CCA) area, and IJV-to-CCA ratios, are then combined with age, body mass index, blood pressure, and heart rate in a dual-modality multilayer perceptron (DM-MLP) designed for clinical tabular data. Its Attribute-Mixing and Case-Mixing operations improved the area under the receiver operating characteristic curve (AUC) by 4–8% compared with ResNet, DenseNet, and Transformer baselines.

In a prospective, multicenter cohort of 349 intensive care unit patients, the DM-MLP detected elevated central venous pressure (CVP; ≥8 mmHg) with AUCs of 0.91 in the internal cohort and 0.87 in the external cohort. Performance remained robust at 7 and 9 mmHg thresholds, while explainability analyses identified IJV-derived indices as the dominant predictors, reinforcing the value of continuously extracted vascular measurements for noninvasive CVP assessment.

The system runs at 32 frames per second on a hospital server, supporting near‑real‑time interpretation. Reported limitations include modest training size, a focus on diagnostic accuracy rather than outcomes, and remaining opacity in semi‑supervised decision pathways. The study was published in Cyborg and Bionic Systems on August 8, 2026, and plans call for larger cohorts, direct prediction of continuous CVP, enhanced interpretability, and interventional trials.

“We designed an ultra‑thin, 128‑element linear‑array ultrasound transducer that can be worn comfortably on the neck. It images the right internal jugular vein (IJV) and the common carotid artery (CCA) – two vessels that reliably reflect central venous pressure because the IJV connects directly to the right atrium without valves,” said Professor Zheng.

“This is not about replacing CVCs in all patients. It is about providing a safe, rapid, and repeatable screening tool for patients in whom catheterization is contraindicated or difficult, and for early bedside identification of elevated CVP to guide timely intervention,” Zheng added.