Automatic Puncture Needle Detection by Image Processing Using Deep Learning and CT Values

In this paper, a needle detection method in CT image for robotized Interventional Radiology (IR) is proposed. There are two stage to detect the needle form. First, the set of 32 square pixels including the needle is abstracts using Convolutional Neural Networks (CNNs). In second stage, pixel-wised form is picked up based on the algorithm to compare CT value on a pixel with its surrounding pixels. 346 sets of CT images were used to train the learner, the evaluation, and evaluate the detection. Finally proposed effectiveness of algorithm is confirmed with simulation.