Endovascular catheters are necessary for state-ofthe- art treatments of life-threatening and time-critical diseases like strokes and heart attacks. Navigating them through the vascular tree is a highly challenging task. We present our preliminary results for the autonomous control of a guidewire through a vessel phantom with the help of Deep Reinforcement Learning. We trained Deep-Q-Network (DQN) and Deep Deterministic Policy Gradient (DDPG) agents on a simulated vessel phantom and evaluated the training performance. We also investigated the effect of the two enhancements Hindsight Experience Replay (HER) and Human Demonstration (HD) on the training speed of our agents. The results show that the agents are capable of learning to navigate a guidewire from a random start point in the vessel phantom to a random goal. This is achieved with an average success rate of 86.5% for DQN and 89.6% for DDPG. The use of HER and HD significantly increases the training speed. The results are promising and future research should address more complex vessel phantoms and the use of a combination of guidewire and catheter.

Flexible steerable needles have the potential to allow surgeons to reach deep targets inside the human body with higher accuracy than rigid needles do. Furthermore, by maneuvering around critical anatomical structures, steerable needles could limit the risk of tissue damage. However, the design of a thin needle (e.g. diameter under 2 mm) with a multi-direction steering mechanism is challenging. The goal of this paper is to outline the design and experimental evaluation of a biologically inspired needle with a diameter under 2 mm that advances through straight and curved trajectories in a soft substrate without being pushed, without buckling, and without the need of axial rotation. The needle design, inspired by the ovipositor of parasitoid wasps, consisted of seven nickel titanium wires and had a total diameter of 1.2 mm. The motion of the needle was tested in gelatin phantoms. Forward motion of the needle was evaluated based on the lag between the actual and the desired insertion depth of the needle. Steering was evaluated based on the radius of curvature of a circle fitted to the needle centerline and on the ratio of the needle deflection from the straight path to the insertion depth. The needle moved forward inside the gelatin with a lag of 0.21 (single wire actuation) and 0.34 (double wire actuation) and achieved a maximum curvature of 0.0184 cmand a deflection-to-insertion ratio of 0.0778. The proposed biologically inspired needle design is a relevant step towards the development of thin needles for percutaneous interventions.

High accuracy and precision in reaching target locations inside the human body is necessary for the success of percutaneous procedures, such as tissue sample removal (biopsy), brachytherapy, and localized drug delivery. Flexible steerable needles may allow the surgeon to reach targets deep inside solid organs while avoiding sensitive structures (e.g. blood vessels). This article provides a systematic classification of possible mechanical solutions for three-dimensional steering through solid organs. A scientific and patent literature search of steerable instrument designs was conducted using Scopus and Web of Science Derwent Innovations Index patent database, respectively. First, we distinguished between mechanisms in which deflection is induced by the pre-defined shape of the instrument versus mechanisms in which an actuator changes the deflection angle of the instrument on demand. Second, we distinguished between mechanisms deflecting in one versus two planes. The combination of deflection method and number of deflection planes led to eight logically derived mechanical solutions for three-dimensional steering, of which one was dismissed because it was considered meaningless. Next, we classified the instrument designs retrieved from the scientific and patent literature into the identified solutions. We found papers and patents describing instrument designs for six of the seven solutions. We did not find papers or patents describing instruments that steer in one-plane on-demand via an actuator and in a perpendicular plane with a pre-defined deflection angle via a bevel tip or a pre-curved configuration.

Purpose Three‐dimensional (3D) printing allows for the fabrication of medical devices with complex geometries, such as soft actuators and robots that can be used in image‐guided interventions. This study investigates flexible and rigid 3D‐printing materials in terms of their impact on multimodal medical imaging. Methods The generation of artifacts in clinical computer tomography (CT) and magnetic resonance (MR) imaging was evaluated for six flexible and three rigid materials, each with a cubical and a cylindrical geometry, and for one exemplary flexible fluidic actuator. Additionally, CT Hounsfield units (HU) were quantified for various parameter sets iterating peak voltage, x‐ray tube current, slice thickness, and convolution kernel. Results We found the image artifacts caused by the materials to be negligible in both CT and MR images. The HU values mainly depended on the elemental composition of the materials and applied peak voltage was ranging between 80 and 140 kVp. Flexible, nonsilicone‐based materials were ranged between 51 and 114 HU. The voltage dependency was less than 29 HU. Flexible, silicone‐based materials were ranged between 60 and 365 HU. The voltage‐dependent influence was as large as 172 HU. Rigid materials ranged between −69 and 132 HU. The voltage‐dependent influence was <33 HU. Conclusions All tested materials may be employed for devices placed in the field of view during CT and MR imaging as no significant artifacts were measured. Moreover, the material selection in CT could be based on the desired visibility of the material depending on the application. Given the wide availability of the tested materials, we expect our results to have a positive impact on the development of devices and robots for image‐guided interventions.

The treatment of cerebro- and cardiovascular diseases requires complex and challenging navigation of a catheter. Previous attempts to automate catheter navigation lack the ability to be generalizable. Methods of Deep Reinforcement Learning show promising results and may be the key to automate catheter navigation through the tortuous vascular tree. This work investigates Deep Reinforcement Learning for guidewire manipulation in a complex and rigid vascular model in 2D. The neural network trained by Deep Deterministic Policy Gradients with Hindsight Experience Replay performs well on the low-level control task, however the high-level control of the path planning must be improved further.