Catheters are increasingly being used to tackle problems in the cardiovascular system. However, positioning precision of the catheter tip is negatively affected by hysteresis. To ensure tissue damage due to imprecise positioning is avoided, hysteresis is to be understood and compensated for. This work investigates the feasibility to model hysteresis with a Long Short-Term Memory (LSTM) network. A bench-top setup containing a catheter distal segment was developed for model evaluation. The LSTM was first tested using four groups of test datasets containing diverse patterns. To compare with the LSTM, a Deadband Rate-Dependent Prandtl-Ishlinskii (DRDPI) model and a Support Vector Regression (SVR) model were established. The results demonstrated that the LSTM is capable of predicting the tip bending angle with sub-degree precision. The LSTM outperformed the DRDPI model and the SVR model by 60.1% and 36.0%, respectively, in arbitrarily varying signals. Next, the LSTM was further validated in a 3D reconstruction experiment using Forward-Looking Optical Coherence Tomography (FL-OCT). The results revealed that the LSTM was able to accurately reconstruct the environment with a reconstruction error below 0.25 mm. Overall, the proposed LSTM enabled precise free-space control of a robotic catheter in the presence of severe hysteresis. The LSTM predicted the catheter tip response precisely based on proximal input pressure, minimizing the need to install sensors at the catheter tip for localization.

In cardiovascular interventions, when steering catheters and especially robotic catheters, great care should be paid to prevent applying too large forces on the vessel walls as this could dislodge calcifications, induce scars or even cause perforation. To address this challenge, this paper presents a novel compliant motion control algorithm that relies solely on position sensing of the catheter tip and knowledge of the catheter's behavior. The proposed algorithm features a data-driven tip position controller. The controller is trained based on a so-called control Long Short-Term Memory Network (control-LSTM). Trajectory following experiments are conducted to validate the quality of the proposed control-LSTM. Results demonstrated superior positioning capability with sub-degree precision of the new approach in the presence of severe rate-dependent hysteresis. Experiments both in a simplified setup as well as in an aortic phantom further show that the proposed approach allows reducing the interaction forces with the environment by around 70%. This work shows how deep learning can be exploited advantageously to avoid tedious modeling that would be needed to precisely steer continuum robots in constrained environments such as the patient's vasculature.

Patch clamping of neurons is a powerful technique used to understand the electrophysiological signals of the brain and advance research into neurological disorders. In in vivo patch clamping, a micropipette is clamped onto the membrane of a neuronal cell body. This technique is difficult and timeconsuming to perform due to the challenges in approaching neurons because of their small size, the absence of visual feedback, and physiologically induced movement caused by heartbeat and breathing. This paper presents a model-based motion compensation algorithm relying solely on electrical bio-impedance (EBI) sensing. The ultimate goal is to cancel out the relative motion between the patch-pipette and the neuron to increase in vivo patch clamping efficiency. In the proposed algorithm, EBI-pipette measurements in response to physiologically induced motions are used to impose on the pipette a motion similar to that of the neuron. The model is based on the assumption that physiological motion can be approximated by a sinusoidal model with three parameters: frequency, phase, and amplitude. The developed compensation algorithm was evaluated in an experimental setup and results yielded a compensation efficiency of (85.5 ± 3.6)%, (81.9 ± 4.0)%, (75.9 ± 1.8)% for artificially imposed motions of 1 Hz, 2 Hz and 3 Hz with an amplitude of 31 µm. The algorithm also demonstrated that it can adjust its motion characterization in real time to changes in amplitude, phase, and also frequency.