Noninvasive Assessment of Arterial Compliance and Lumen Pressure in Human Carotid Arteries Using Physics-Informed Neural Networks and Ultrasound Imaging: A Clinical Feasibility Study

Arterial stiffness is an independent predictor of cardiovascular mortality, with localized compliance variations providing key insights for vascular disease assessment. In our previous studies, we implemented a PINN-based inverse approach integrating Pulse Wave Imaging (PWI) and Vector Flow Imaging (VFI) data with 1D PDEs of pulse wave propagation and was trained to minimize mismatch between measured data and physics constraints for in silico and phantom-mimicking stenotic vessels. This study evaluates the performance of the previously-reported PINN framework for noninvasive, spatially resolved estimation of arterial compliance and lumen pressure from the high-frame-rate ultrasound (PWI and VFI) data in a clinical setting. The framework was applied on common carotid artery high-frequency ultrasound frames (3 kHz frame rate) in five subjects: one healthy and four with carotid stenosis (low to medium occlusion). The proposed model estimated spatially varying compliance in all subjects: healthy subjects showed uniform compliance, while stenotic subjects exhibited focal compliance variation correlating with plaques/occlusion on ultrasound B-mode images. The model reconstructed wall motion (% difference < 0.2% in healthy, < 0.48% in carotid stenosis patients) and flow velocity (% difference < 0.2% in healthy). In stenotic cases, the model adapted to unreliable flow data by using the wall motion data only. The computation time ranged from 20 minutes to 1.5 hour depending on the spatial lateral resolution; lateral resolution of 16-elements data (instead of full 128-elements data of L7-4 transducer) considered for healthy cases required 20 minutes runtime, whereas, for the stenotic cases, 1.5 hours was needed due to the required full resolution (all 128-elements data). In conclusion, this study demonstrates the effectiveness of the PINN-based framework for non-invasive, patient-specific mapping of localized arterial compliance variation as well as luminal pressure. These findings presented in this feasibility study support clinical translation for non-invasive atherosclerotic plaque risk evaluation and personalized cardiovascular assessment.