Tri-Axial Force Sensor in a Soft Catheter Using Fiber Bragg Gratings for Endoscopic Submucosal Dissection

• Published in: IEEE sensors journal 2023-10, Vol.23 (20), p.24626-24636
• Main Authors: Ben Hassen, Ramzi, Lemmers, Arnaud, Delchambre, Alain
• Format: Article
• Language: English
• Subjects: Axial forces; Bragg gratings; Catheters; Contact force; Dissection; Electrostatic discharges; Endoscopic submucosal dissection (ESD); Endoscopy; fiber bragg gratings (FBGs); Fiber gratings; Force; force calibration; force sensor; Manganese; Neural networks; Nickel titanides; Nonlinearity; Prototyping; regression model; Regression models; Robot sensing systems; Root-mean-square errors; Sensors; Support vector machines; therapeutic endoscopy; training in ESD; Tubes
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2023.3313172

Endoscopic submucosal dissection (ESD) is an advanced endoscopic technique with renowned clinical benefits but still a challenging procedure. The lack of force-feedback leading to insufficient or excessive contact force between the tip of the knife and the tissue, i.e., ineffective treatment or dangerous perforation, makes ESD require a high level of expertise and dexterity to master it, especially for trainees. In this article, we propose to enhance the training in ESD by integrating fiber Bragg gratings (FBGs) as three degrees-of-freedom optical force sensors into the polymer catheter of the electrosurgical knife aiming to measure {F}_{x} , {F}_{y} , and {F}_{z} . Three FBGs are placed circumferentially to the Section of the catheter using nitinol tubes and a two-point pasting method. A force calibration test bench was specifically designed to calibrate the force sensor in 30 3-D spatial directions that cover most of its use cases. Nonlinear regression models were implemented to tackle the nonlinearities between the wavelength shifts of the FBGs and the forces applied, inherent to prototyping errors and nonlinearity of the soft material. A hybrid model made of mono- and bi-layered neural networks (NNs) for the prediction of {F}_{x} and {F}_{y} and a support vector regression (SVR) for the prediction of {F}_{z} was built and showed root-mean-square error (RMSE) along transverse directions ( \textit {XY} ) less than 3% of the full scale [−500; 500] mN and RMSE less than 10% along the axial direction ( {Z} ). These models were also verified in dynamic conditions. The results are promising and satisfy all the technical requirements.