Central venous pressure (CVP) is a key indicator of right ventricular preload, but measurement with a central venous catheter (CVC) is invasive, time-consuming, and unsuitable for bedside monitoring. We developed and clinically evaluated an artificial intelligence (AI)-enabled wearable ultrasound device for noninvasive estimation of elevated CVP by imaging the internal jugular vein (IJV) and common carotid artery (CCA) in acute and critical patients. In this prospective multi-center clinical study, 349 patients admitted to intensive care units (ICUs) at 2 tertiary hospitals underwent neck vascular imaging with the wearable ultrasound device, while CVP was measured simultaneously via CVC as the reference standard. The dual-decoder spatiotemporal attention network (DSTA-Net) segmentation model automatically quantified IJV and CCA cross-sectional areas, and the dual-modality multilayer perceptron (DM-MLP) integrated these vascular indices with basic clinical parameters to predict elevated CVP (≥8 mmHg). DSTA-Net showed excellent agreement and strong correlation with expert manual measurements while achieving superior segmentation accuracy compared with the baseline segmentation model. Meanwhile, the proposed DM-MLP improved prediction performance by 4% to 8% over conventional baseline architectures, achieving AUCs of 0.91 (internal test set) and 0.87 (external test set). These findings indicate that an AI-integrated wearable ultrasound device can provide accurate, noninvasive bedside assessment of CVP and may offer a promising alternative to invasive catheters for hemodynamic monitoring in acute and critical care settings.