Computer Vision News 34 Congrats, Doctor Maria! Ultrasound imaging is a known imaging technique in medical imaging, allowing for real-time data acquisition without the use of ionizing radiation. However, ultrasound imaging presents several challenges, among which the inter-patient and inter-device variability, Maria Tirindelli has obtained her PhD last week at TUM, at the chair of Computer Aided Medical Procedure and Augmented Reality (CAMP) under the supervision of Nassir Navab. Since last year, Maria is working as a Research Scientist at ImFusion GmbH. Congrats, Doctor Maria! variability, noise, and artifacts. Additionally, the acquisition process strongly depends on the operator, making the procedure reproducibility low. To address these issues, robotic ultrasound has been proposed in the literature, to automate the acquisition of ultrasound data. Robotic ultrasound presents several advantages, as it reduces the reliance on the operator's expertise, it can guide novice radiologists in the acquisition process, and it relieves the operators from the task of manipulating the ultrasound probe. Furthermore, robotic ultrasound can be a valuable tool to enlarge the diagnostics and treatment reachability to remote areas, where the presence of medical staff can be limited. However, automatic data acquisition and interpretation remain a challenge. This dissertation addresses the challenges of data interpretation and trajectory optimization for optimal ultrasound quality in robotic ultrasound. In the first work of the dissertation, we propose a new method for automatic ultrasound acquisition and vertebral level identification for spinal injection. Specifically, we propose a setup consisting of a robotic arm, where a force sensor and an ultrasound probe are mounted. We then program the robot to move along the spine on the patient's back, while ultrasound and force data are acquired. Thanks to the utilized force control, we can then define a model for patient-robot arm interactions, to extract a force signal where vertebrae position can be identified along the spine. We then use a Convolutional Neural Network to extract vertebrae positions from the ultrasound data and we fuse the information, to guide the robot to the correct target.
RkJQdWJsaXNoZXIy NTc3NzU=