Session: 07-10-02: Medical Robotics, Rehabilitation, and Surgery

Paper Number: 193462

Neural Network-Driven Adaptive Needle Steering Controller: Out-of-Plane Deflection Control in HDR Brachytherapy Applications 

Introduction: In high-dose-rate (HDR) brachytherapy procedures, clinical efficacy depends on the precise positioning of the needle assisted by robotic systems. To improve the patient's safety and target positioning accuracy, advanced controller design for steerable needling system is one of the crucial objectives. However, uncertainties within the system and complex non-holonomic constraints during needle insertion inside tissue, offer significant challenges for precise needle actuation. These modeling uncertainties can compromise the effectiveness of control strategies, potentially resulting in suboptimal or unstable needle trajectories.

Contributions: The major contributions of this work are: (1) Design and development of an adaptive needle steering controller using neural network (NN) architecture while controlling out-of-plane deflection, (2) Controller validation through simulation studies in MATLAB while conducting comprehensive comparison study with the existing needle steering techniques.

Methodology:

Needle Kinematic Model: This study incorporated the differential kinematic model of the needing system using unicycle model with Z-Y-X fixed angle configuration. The actuation of the needle was considered from the needle base, using a combination of two control inputs: angular velocity around needle base, and insertion velocity along needle axis. In this study, we have assumed that the needle has negligible torsional compliance, resulting in a 2-DOF needling system. The objective of the designed control scheme was to limit the needle-tip deflection from a desired plane. For this purpose, the controller was designed for y-z plane with needle insertion along z axis. Correspondingly, the system state equation was designed using the time derivative of deflection state (x position), and the angles around y and z axis. Notably, the insertion velocity was treated as constant while angular velocity was considered as a primary control input to achieve the desired performance.

Controller Design: This work focuses on development of an NN-based adaptive controller for needle steering, incorporating non-holonomic constraints during needle insertion inside the tissue. The novelty of this work lies in the development of customized weight update laws for the NN architecture, while ensuring the closed loop analysis. To obtain the model’s weight update laws, Lyapunov stability analysis was conducted with the involvement of sliding surface concept. The NN architecture was designed based on one hidden layer with 20 neurons and three inputs (deflection, time derivative of deflection, and 

integration of deflection). The output layer of the NN architecture provides the proper needle steering control signal.

Simulation Results: In this study, the proposed control scheme was developed and simulated in MATLAB environment. The proposed control scheme achieved the finite convergence of the deflection state (x) within an insertion depth of 4 to 6 cm, with RMSE of 0.52 mm. Meanwhile, the existing control schemes showed the reduced performance as, RMSE in the range of 1.02 to 1.14 mm. Notably, in needle interventional procedure, precise regulation of needle placement within minimal insertion depth is crucial, whereas existing controllers demonstrated higher insertion depth (range: 9 to 10 cm) compared to our proposed scheme.

Conclusions: This study demonstrated the design of an NN-based adaptive needle steering scheme to control out-of-plane deflection during needle interventional procedures. The comprehensive comparative analysis highlighted superior performance compared to existing needle steering techniques (state feedback, observer-based, and adaptive neuro fuzzy inference system-based control). However, the designed controller was validated using simulation studies, real-time experimental validation will be conducted in future work to show its efficacy.

Presenting Author: Kaushik Halder SUNY Upstate Medical University

Presenting Author Biography: Kaushik Halder received the B.Sc in Physics Hons. and B.Tech. degree in electrical engineering
from University of Calcutta, West Bengal, India. He received M.Tech. degree in Electrical
Engineering with the specialization in Advanced Control and Measurement System from
Applied Physics Department, University of Calcutta, West Bengal, India. He received the PhD
degree in electrical engineering from the Indian Institute of Technology Roorkee, Uttarakhand,
India, in 2025. He is currently serving as a Research Scientist in the department of Radiation
Oncology at SUNY Upstate Medical University, NY, USA. His research interests include
medical robotics, intelligent control techniques, adaptive control and system identification.

Authors:

Kaushik Halder SUNY Upstate Medical University
M. Felix Orlando IIT Roorkee
Parsaoran Hutapea Temple University
Jeffrey A. Bogart SUNY Upstate Medical University
Tarun Podder SUNY Upstate Medical University