A Deep Learning-Assisted Ultrasound Imaging Workflow for Patient-Specific Carotid Artery Modeling and Phantom Fabrication
Affiliation Type:
Academia
Keywords:
Carotid Flow Dynamics, MedSAM, Digital Twin
Abstract:
This study developed an integrated pipeline combining in vivo ultrasound imaging, phantom fabrication, and hemodynamic simulations for carotid artery analysis to assess stroke risk. A custom ultrasound scanning apparatus with three universal joints stabilized the probe, reducing free-hand misalignment. The MedSAM model was enhanced with adaptive prompting and a bifurcation detection module for automated lumen segmentation, improving Dice scores by 0.102. Patient-specific 3D carotid models were reconstructed, and arterial stiffness and blood flow were quantified using shear wave imaging and Doppler ultrasound. PVA hydrogel phantoms replicating in vivo stiffness were fabricated using a 3D-printed mold. Fluid-structure interaction simulations matched Doppler ultrasound results, showing a 7:3 ICA/ECA flow ratio and flow separation at the bifurcation. The phantom had an average wall thickness of 2.6 mm and Young’s modulus of 77 kPa. Ongoing work focuses on reproducibility and pathological condition modeling.
