Loading…

A markerless beam’s eye view tumor tracking algorithm based on unsupervised deformable registration learning framework of VoxelMorph in medical image with partial data

•Propose a markerless BEV tumor radiotherapy tracking algorithm.•Requires no additional equipment; free of secondary trauma and additional dose.•Achieve the alignment of MLC occlusion images with template matching method.•End-to-end unsupervised framework for non-rigid registration of partial image...

Full description

Saved in:
Bibliographic Details
Published in:Physica medica 2023-01, Vol.105, p.102510-102510, Article 102510
Main Authors: Zhong, Jiajian, Huang, Taiming, Qiu, Minmin, Guan, Qi, Luo, Ning, Deng, Yongjin
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Propose a markerless BEV tumor radiotherapy tracking algorithm.•Requires no additional equipment; free of secondary trauma and additional dose.•Achieve the alignment of MLC occlusion images with template matching method.•End-to-end unsupervised framework for non-rigid registration of partial image data.•Processes inferior quality MV X-ray images with high noise level and low contrast. To propose an unsupervised deformable registration learning framework-based markerless beam's eye view (BEV) tumor tracking algorithm for the inferior quality megavolt (MV) images with occlusion and deformation. Quality assurance (QA) plans for thorax phantom were delivered to the linear accelerator with artificially treatment offsets. Electronic portal imaging device (EPID) images (682 in total) and corresponding digitally reconstructed radiograph (DRR) were gathered as the moving and fixed image pairs, which were randomly divided into training and testing set in a ratio of 0.7:0.3 to train a non-rigid registration model with Voxelmorph. Simultaneously, 533 pairs of patient images from 21 lung tumor plans were acquired for tumor tracking investigation to offer quantifiable tumor motion data. Tracking error and image similarity measures were employed to evaluate the algorithm’s accuracy. The tracking algorithm can handle image registration with non-rigid deformation and losses ranging from 10 % to 80 %. The tracking errors in the phantom study were below 3 mm in about 86.8 % of cases, and below 2 mm in about 80 % of cases. The normalized mutual information (NMI) changes from 1.182 ± 0.024 to 1.198 ± 0.024 (p 
ISSN:1120-1797
1724-191X
DOI:10.1016/j.ejmp.2022.12.002