Towards expert-level autonomous carotid ultrasonography with large-scale learning-based robotic system.

Area of Science:

• Robotics
• Medical Imaging
• Artificial Intelligence

Background:

• Carotid ultrasound demands specialized skills due to anatomical variations, leading to shortages and inconsistencies.
• Previous automation efforts lacked generalizability and failed in real-world clinical settings.

Purpose of the Study:

• To develop a fully learning-based autonomous robot for carotid ultrasound.
• To achieve human-expert-level performance and overcome limitations of prior automation attempts.

Main Methods:

• Implemented a unified imitation learning framework for skill acquisition.
• Utilized a large-scale expert demonstration dataset (247,000 samples) for embodied foundation models.
• Developed a comprehensive scanning protocol for complete anatomical coverage.

Main Results:

• Achieved over 90% success rates in clinical validation.
• Demonstrated expert-level accuracy and up to 5.5x higher reproducibility.
• Showcased strong generalization across diverse, unseen populations.

Conclusions:

• Large-scale deep learning enables autonomous, high-precision ultrasonography.
• UltraBot represents a promising advancement for clinical carotid ultrasound practice.