A Recurrent Neural Network Approach to Roll Estimation for Needle Steering
This addresses the challenge of accurate needle steering for medical procedures, but it is incremental as it applies an existing neural network method to a specific domain problem.
The paper tackled the problem of estimating needle tip orientation for steerable needles in minimally-invasive therapy, proposing a model-free LSTM neural network approach that achieved significantly lower targeting errors compared to an Extended Kalman Filter in gelatin and ex vivo ovine brain tissue.
Steerable needles are a promising technology for delivering targeted therapies in the body in a minimally-invasive fashion, as they can curve around anatomical obstacles and hone in on anatomical targets. In order to accurately steer them, controllers must have full knowledge of the needle tip's orientation. However, current sensors either do not provide full orientation information or interfere with the needle's ability to deliver therapy. Further, torsional dynamics can vary and depend on many parameters making steerable needles difficult to accurately model, limiting the effectiveness of traditional observer methods. To overcome these limitations, we propose a model-free, learned-method that leverages LSTM neural networks to estimate the needle tip's orientation online. We validate our method by integrating it into a sliding-mode controller and steering the needle to targets in gelatin and ex vivo ovine brain tissue. We compare our method's performance against an Extended Kalman Filter, a model-based observer, achieving significantly lower targeting errors.