Deep Multimodal Imitation Learning-based Framework for Robot-assisted Medical Examination

Researchers from the Bristol Robotics Laboratory have proposed a novel framework for robot-assisted medical examination, specifically for the initial registration in artery scanning, using deep multimodal imitation learning, compliant control, and trajectory optimization. This framework aims to improve the dexterity and accuracy of robots in performing complex medical tasks, such as ultrasound examination, by leveraging deep learning techniques and real-time feedback from the patient. According to the researchers, the proposed approach significantly improves the success rate of autonomous ultrasound scanning from 75% to 90% while reducing the completion time.

Key Takeaways:

• The proposed framework integrates deep multimodal imitation learning, compliant control, and trajectory optimization to improve the dexterity and accuracy of robots in medical examinations.
• The framework uses a deep imitation learning model that fuses RGB and ultrasound images, contact force, and proprioceptive data to predict the desired motion and contact force.
• The compliant controller in Cartesian space tracks the desired trajectory and force, while a trajectory optimization planner smooths the trajectory and ensures safety.
• The framework was evaluated on both Phantom and human subjects, demonstrating significant improvements in success rate and completion time.
• The authors propose a unified framework for robot-assisted medical examination, enhancing the autonomy and accuracy of robots in complex medical tasks.
• Researchers Chenguang Yang, Weiyong Si, Ning Wang, and Rebecca Harris contributed to the study, with Yang serving as the lead researcher at the Bristol Robotics Laboratory.

Statistics:

• The proposed approach improved the success rate of autonomous ultrasound scanning from 75% to 90%.
• The completion time was reduced, although specific details are not provided in the news report.
• The framework uses a deep multimodal imitation learning model that fuses RGB and ultrasound images, contact force, and proprioceptive data.
• The research was conducted by the Bristol Robotics Laboratory and published in IEEE Transactions On Industrial Electronics.

Sources:

• Deep Multimodal Imitation Learning-based Framework for Robot-assisted Medical Examination. Ieee Transactions On Industrial Electronics, 2025.
• Ieee Transactions On Industrial Electronics can be contacted at: Ieee-inst Electrical Electronics Engineers Inc, 445 Hoes Lane, Piscataway, NJ 08855-4141, USA.
• Institute of Electrical and Electronics Engineers - www.ieee.org/
• Ieee Transactions On Industrial Electronics - ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=41