Multimodal Point Cloud Registration Based on Adaptive Feature Region in Radiotherapy Guidance

Abstract

In view of the positioning error existing tumor radiotherapy, we have adopted the radiotherapy guidance system to assist accurate calibration. The following research has be done to solve this problem: an optical three-dimensional scanner is used to accurately register the body surface images with the CT images before radiotherapy to guide the positioning of radiotherapy. Firstly, the two types of pre-processed image information are converted into point cloud data storage with similar format, and the problem of guiding position and multimodal registration are transformed into the problem of three-dimensional point clouds registration. Then, the initial estimation correspondence of the point clouds and the initial transformation matrix are calculated by coarse registration algorithm. Finally, Iterative Closest Points (ICP) algorithm with dynamic threshold is used for fine registration. Among them, an adaptive feature region extraction method is proposed to extract the most obvious feature region in the image for registration, which effectively simplifies the point cloud data and improves the accuracy of registration. The simulation results prove the effectiveness of this method, which has certain reference value for guiding positioning in radiotherapy. A new multi-modal image registration method is proposed. On this basis, it can be further extended and applied on some other problems.