3D Robotic Ultrasound

For diagnosis and treatment planning in orthopedics, e.g. total knee arthroplasty, 3D imaging is often required. Today’s standard for 3D image acquisition is CT or MRI. Both techniques cause high costs and CT emits harming radiation in addition. In contrast, ultrasound is commonly available at chairside, comes at low costs and does not expose the patient to radiation. Interpretation of ultrasound images however is a difficult task, in particular for volumetric images. Sonographer require years of experience to reliably detect the bone surface. Fast and precise diagnostics, e.g. of cruciate ligament rupture or periimplantitis, thus depends on automatic processing. We investigate the potential of robotic ultrasound scanning.

For image acquisition the robot moves along the region of interest while a contact force between the skin and the probe must be maintained. Challenges include robot control strategies as well as image processing for an accurate bone segmentation.

Latest research involves customized machine learning architectures for leveraging the spatial information: The transformer architecture, originally developed for natural language processing tasks, proofs to be a promising alternative to the convolutional neural network.