Neural Network Modeling of Maximum Insertion Force of Bevel-Tip Surgical Needle

Recently there has been a growing interest to develop innovative surgical needles for percutaneous interventional procedures. Needles are commonly used to reach target locations inside of the body for various medical interventions. The effectiveness of these procedures depends on the accuracy with which the needle tips reach the targets, such as a biopsy procedure to assess cancerous cells and tumors. One of the major issues in needle steering is the force during insertion, also known as the insertion (penetration) force. The insertion force causes tissue damage as well as tissue deformation. It has been well studied that tissue deformation causes the needle to deviate from its target thus causing an ineffective procedure. Simulation of surgical procedures provides an effective method for a robot-assisted surgery for pre- and intra-operative planning. Accurate modeling of the mechanical behavior on the interface of surgical needles and organs, specifically the insertion force, has been well recognized as a major challenge. Overcoming such obstacle by development of robust numerical models will enable realistic force feedback to the user during surgical simulation. This study investigates feasibility of predicting the insertion force of bevel-tip needles based on experimental data using neural network modeling. Simulation of the proposed neural network model is performed using Kera's Python Deep Learning Library with TensorFlow as a backend. The insertion forces of needles with different bevel-tip angles in gel tissue phantom are measured using a specially designed automated needle insertion test setup. Input-output datasets are generated where the inputs are defined as bevel-tip angles and gel tissue phantom stiffness, and the output is defined as the insertion force. A properly trained neural network then maps the input data to the output data and the input-output dataset is supplied to train a neural network. Its performance is then evaluated using different and unseen input-output dataset. This paper shows that the proposed neural network model accurately predicts the insertion force.