Abstract—In contemporary brachytherapy procedure, needle placement at desired location is challenging due to a variety of reasons. We have designed a robot-assisted brachytherapy system to improve needle placement and seed delivery. In this paper, we have used neural network (NN) for predicting insertion force during prostate brachytherapy. The NN controller computes control inputs required for optimizing the robotic system. To verify efficacy of the control system we used in-vivo motion and force measurements during actual brachytherapy needle insertion while radioactive seeds were implanted in the prostate gland, as a real-time controller input signal. Both force prediction and force tracking processes are investigated. Information about insertion force values are used to adjust other insertion parameters like insertion velocity or acceleration in order to minimize the insertion force. Index Terms — Robotic brachytherapy, neural network controller, force prediction, adaptive control system. I. INTRODUCTION OW-DOSE rate (LDR) prostate brachytherapy is a method of delivering radiotherapy by implanting radioactive sources into and around the prostate gland [1].This procedure is performed under the guidance of transrectal ultrasound (TRUS) images. In traditional brachytherapy procedures, the needles are inserted transperineally under the guidance of transrectal ultrasound images [2]. Both the needle and the ultrasound are operated manually. The seeds are deposited using a manual applicator. In order to increase accuracy of needle placement and seed delivery, we developed a fully automated robotic system for prostate brachytherapy. Using robotic approach we are able to record needle insertion forces and motion trajectories measured during actual brachytherapy needle insertion while implanting radioactive seeds in the prostate gland [3], [4]. These insertion forces are significantly responsible for needle deviation from the desired trajectory and target movement. Proper selection of the translational and rotational velocities may reduce the tissue deformation and target movement by reducing insertion force, [5]-[7]. In the current study, we used a neural network predictive controller to predict insertion force in order to achieve real- Corresponding author: Ivan Buzurovic, Thomas Jefferson University, Philadelphia, PA 19107, email: ivan.buzurovic @jeffersonhospital.org. time adaptive needle control. We tested whether our system is capable of track significant changes of the insertion force and whether it is able to predict that force. The implemented optimization algorithm then computes control signals that optimize future system performances. Measurement and prediction of insertion force and needle fracture force using experimental approach was investigated in [8], [9]. Prediction of how the skin deforms upon insertion by microneedles is described in [10]. Artificial neural networks (ANN) are applied in early detection of prostate cancer [11], [12]. In miniature robotic surgical systems ANN are used in conjunction with real-time visual feedback to “learn” the inverse system dynamics and control the manipulator endpoint trajectory, [13]. Real-time control of a robot arm, using recorded neurons in the motor cortex together with mathematical transformations including neural networks, is presented in [14]. Overview of the ANN applications in many disciplines, especially in medicine can be found in [15]. In the [16], authors solved problem of achieving high accuracy positioning of a medical robot using neural network for the patient positioning system. Another robotic system for heart surgery which uses a recurrent neural network with adaptive internal states is described in [17]. In [18], it is shown that the neural network demonstrated robust adaptability to all of the observed breathing patterns while the linear filter failed in a significant percentage of cases. This comprehensive literature survey provides us with the motivation for developing a neural network predictive controller to predict insertion force and to design adaptive needle control for improving the surgical procedures. II. SYSTEM DESCRIPTION We developed a 16 degree of freedom (DOF) robot-assisted brachytherapy system, [2], [19]. This robotic system is divided into three subsystems: cart, supporting platform, and surgery modules, with 6, 3 and 7 DOF, respectively. The supporting platform connects the surgery module to the cart. The surgery module consists of 2DOF ultrasound probe driver and 5DOF needling module (Fig. 1). The ultrasound (US) module can be translated and rotate independently by two DC servomotors fitted with high-resolution optical encoders and gearboxes. In this current study, we investigate 5DOF-needling module, Ivan Buzurovic, Tarun K. Podder, and Yan Yu Department of Radiation Oncology, Kimmel Cancer Center (NCI-designated), Thomas Jefferson University, Philadelphia, PA 19107, USA Email: ivan.buzurovic@jeffersonhospital.org Force Prediction and Tracking for Image-guided Robotic System using Neural Network Approach L 978-1-4244-2879-3/08/$25.00 ©2008 IEEE 41