Prediction of puncturing events through LSTM for multilayer tissue.

Abstract

Recognizing penetration events in multilayer tissue is critical for many biomedical engineering applications, including surgical procedures and medical diagnostics. This paper presents a unique method for detecting penetration events in multilayer tissue using Long Short-Term Memory (LSTM) networks. LSTM networks, a form of recurrent neural network (RNN), excel at analyzing sequential data because of their ability to hold long-term dependencies. The suggested method collects time-series insertion force data from sensors integrated from a 1-DOF prismatic robot as it penetrates tissue. This data is then processed by the LSTM network, which has been trained to recognize patterns indicating penetration events through various tissue layers. The effectiveness of this approach is validated through experimental setups, demonstrating high accuracy and reliability in detecting penetration events. This technique offers significant improvements over traditional methods, providing a non-invasive, real-time solution that enhances the precision and safety of medical procedures involving multilayer tissue interaction.