Feasibility Study on Using Supervised Deep-Learning to Model Edge-FBG Shape Sensors

This study is part of the MIRACLE project, an ongoing project of the Department of Biomedical Engineering at the University of Basel. The main objective of the project MIRACLE, short for Minimally Invasive Robot-Assisted Computer-guided LaserosteotomE, is to develop an innovative robotic approach to perform a contact-free bone surgery with laser light. The laser beam will be guided through a flexible endoscope. The cut performed by the laser light has several advantages over the cut made with a conventional saw such as higher precision, quick recovery time of the patient, less trauma, and potential to cut bone pieces in angles and shapes that are not possible with traditional methods.
Real-time feedback on the exact shape and location of the flexible endoscope is required to determine its position inside the body of the patient. Due to the flexibility of the endoscope, lack of line of sight, and magnetic interference, the existing tracking technologies such as electromagnetic sensors and optical markers are not suitable. Thus, it is necessary to develop a new navigation system to monitor the shape and the tip position of the endoscope. Fiber-based shape sensing is a promising approach that has received recent attention due to its small size, biocompatibility, high sensitivity, and immunity to electromagnetic noise.
One of the most recent types of fiber shape sensors is based on edge-FBGs, a single-mode optical fiber which has on the edge of its core sites, Bragg gratings, changing the refractive index. In edge-FBG sensors, the shape information is hidden in the amplitude of the Bragg wavelength. Therefore, the measured change of these amplitudes could be used to calculate the curvature of the fiber. Unfortunately, an accurate mathematical model describing their behavior is unknown yet, as the sensor's signal is often affected by other bending-related phenomena.
Machine learning techniques have the potential to directly relate the spectrum of the reflected signal to the spatial shape of the sensor and distinguish between the main signal and the noise caused by undesired bending related phenomena. This work was focused on preparing the experimental setup and finding an exemplary neural network to reconstruct the fiber's shape and position, given spectral data of the edge-FBG sensor. The reconstruction of the fiber’s shape using a neural network is further called modeling an edge-FBG sensor.