Transformer-Based Local Feature Matching for Multimodal Image Registration
Authors:
Remi Delaunay,
Ruisi Zhang,
Filipe C. Pedrosa,
Navid Feizi,
Dianne Sacco,
Rajni Patel,
Jayender Jagadeesan
Abstract:
Ultrasound imaging is a cost-effective and radiation-free modality for visualizing anatomical structures in real-time, making it ideal for guiding surgical interventions. However, its limited field-of-view, speckle noise, and imaging artifacts make it difficult to interpret the images for inexperienced users. In this paper, we propose a new 2D ultrasound to 3D CT registration method to improve sur…
▽ More
Ultrasound imaging is a cost-effective and radiation-free modality for visualizing anatomical structures in real-time, making it ideal for guiding surgical interventions. However, its limited field-of-view, speckle noise, and imaging artifacts make it difficult to interpret the images for inexperienced users. In this paper, we propose a new 2D ultrasound to 3D CT registration method to improve surgical guidance during ultrasound-guided interventions. Our approach adopts a dense feature matching method called LoFTR to our multimodal registration problem. We learn to predict dense coarse-to-fine correspondences using a Transformer-based architecture to estimate a robust rigid transformation between a 2D ultrasound frame and a CT scan. Additionally, a fully differentiable pose estimation method is introduced, optimizing LoFTR on pose estimation error during training. Experiments conducted on a multimodal dataset of ex vivo porcine kidneys demonstrate the method's promising results for intraoperative, trackerless ultrasound pose estimation. By mapping 2D ultrasound frames into the 3D CT volume space, the method provides intraoperative guidance, potentially improving surgical workflows and image interpretation.
△ Less
Submitted 25 April, 2024;
originally announced April 2024.
Design and Modeling of a Smart Torque-Adjustable Rotary Electroadhesive Clutch for Application in Human-Robot Interaction
Authors:
Navid Feizi,
S. Farokh Atashzar,
Mehrdad R. Kermani,
Rajni V. Patel
Abstract:
The increasing need for sharing workspace and interactive physical tasks between robots and humans has raised concerns regarding safety of such operations. In this regard, controllable clutches have shown great potential for addressing important safety concerns at the hardware level by separating the high-impedance actuator from the end effector by providing the power transfer from electromagnetic…
▽ More
The increasing need for sharing workspace and interactive physical tasks between robots and humans has raised concerns regarding safety of such operations. In this regard, controllable clutches have shown great potential for addressing important safety concerns at the hardware level by separating the high-impedance actuator from the end effector by providing the power transfer from electromagnetic source to the human. However, the existing clutches suffer from high power consumption and large-weight, which make them undesirable from the design point of view. In this paper, for the first time, the design and development of a novel, lightweight, and low-power torque-adjustable rotary clutch using electroadhesive materials is presented. The performance of three different pairs of clutch plates is investigated in the context of the smoothness and quality of output torque. The performance degradation issue due to the polarization of the insulator is addressed through the utilization of an alternating current waveform activation signal. Moreover, the effect of the activation frequency on the output torque and power consumption of the clutch is investigated. Finally, a time-dependent model for the output torque of the clutch is presented, and the performance of the clutch was evaluated through experiments, including physical human-robot interaction. The proposed clutch offers a torque to power consumption ratio that is six times better than commercial magnetic particle clutches. The proposed clutch presents great potential for developing safe, lightweight, and low-power physical human-robot interaction systems, such as exoskeletons and robotic walkers.
△ Less
Submitted 16 October, 2022;
originally announced October 2022.
