ReUINet: A fast GNL distortion correction approach on a 1.0 T MRI‐Linac scanner

• DOI: 10.1002/mp.14861
• Publication Date: 2021-03-25T01:46:05Z
• Authors: Shanshan Shan; Mao Li; Mingyan Li; Fangfang Tang; Stuart Crozier; Feng Liu
• Keywords: Interpolation; Distortion (music)

Purpose The hybrid system combining a magnetic resonance imaging (MRI) scanner with linear accelerator (Linac) has become increasingly desirable for tumor treatment because of excellent soft tissue contrast and nonionizing radiation. However, image distortions caused by gradient nonlinearity (GNL) can have detrimental impacts on real‐time radiotherapy using MRI‐Linac systems, where accurate geometric information tumors is essential. Methods In this work, we proposed deep convolutional neural network‐based method to efficiently re cover u ndistorted i mages ( ReUINet ) guidance. , based the encoder‐decoder structure, was created learn relationship between undistorted images distorted images. pretrained tested publically available brain MR dataset acquired from 23 volunteers. Then, transfer learning adopted implement model (i.e., network optimal weights) experimental three‐dimensional (3D) grid phantom in‐vivo pelvis datasets 1.0 T Australian system. Results Evaluations (768 slices) data (88 showed that achieved improvement over 15 times 45 computational efficiency in comparison standard interpolation GNL‐encoding methods, respectively. Moreover, qualitative quantitative results demonstrated provided better correction than method, comparable performance compared approach. Conclusions Validated simulation results, promise obtaining implementation MRI‐guided radiotherapy.