Robotic Respiratory Liver Motion Compensation for Stable Ultrasound Imaging of Small Lesions

Real-time ultrasound-guided percutaneous liver biopsy plays a crucial role in diagnosing early liver cancer. However, patient respiration can cause liver drift, causing small target lesions to deviate from the ultrasound imaging plane and become invisible, complicating accurate punctures. To address this, we introduce an automatic robotic respiratory liver motion compensation method, which relies on real-time estimation of liver motion to maintain the stability between the ultrasound imaging plane and the lesion observation plane. We propose an end-to-end Respiratory liver Motion Estimation network, termed ReME-Net. This network effectively utilizes the difference features between the current and target ultrasound frames, combined with the sequential channel-spatial attention mechanism, to accurately estimate the six-degree-of-freedom respiratory liver motion in real time, without any tracking device. We employ a hybrid force-position control framework with an adaptive variable admittance controller to ensure stable probe-skin contact while compensating for liver motion, facilitating stable imaging of small lesions. On this basis, a robotic ultrasound imaging system is developed and experiments on a liver motion simulation platform demonstrate successful tracking with mean errors of 3.05 mm in translation, 2.52° in rotation, and 0.25 N in contact force, while periodic liver motion has a translation amplitude of 19.72 mm and a rotation amplitude of 17.32°. This method promises stable imaging of small liver lesions during breathing, which can improve early tumor puncture accuracy, helping medical professionals.