Premium
SU‐E‐J‐08: A Hybrid Three Dimensional Registration Framework for Image‐Guided Accurate Radiotherapy System ARTS‐IGRT
Author(s) -
Wu Q,
Pei X,
Cao R,
Hu L,
Wu Y
Publication year - 2014
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4888059
Subject(s) - image registration , image guided radiation therapy , voxel , computer science , artificial intelligence , computer vision , medical imaging , mutual information , similarity (geometry) , software , margin (machine learning) , computation , image (mathematics) , algorithm , programming language , machine learning
Purpose: The purpose of this work was to develop a registration framework and method based on the software platform of ARTS‐IGRT and implement in C++ based on ITK libraries to register CT images and CBCT images. ARTS‐IGRT was a part of our self‐developed accurate radiation planning system ARTS. Methods: Mutual information (MI) registration treated each voxel equally. Actually, different voxels even having same intensity should be treated differently in the registration procedure. According to their importance values calculated from self‐information, a similarity measure was proposed which combined the spatial importance of a voxel with MI (S‐MI). For lung registration, Firstly, a global alignment method was adopted to minimize the margin error and achieve the alignment of these two images on the whole. The result obtained at the low resolution level was then interpolated to become the initial conditions for the higher resolution computation. Secondly, a new similarity measurement S‐MI was established to quantify how close the two input image volumes were to each other. Finally, Demons model was applied to compute the deformable map. Results: Registration tools were tested for head‐neck and lung images and the average region was 128*128*49. The rigid registration took approximately 2 min and converged 10% faster than traditional MI algorithm, the accuracy reached 1mm for head‐neck images. For lung images, the improved symmetric Demons registration process was completed in an average of 5 min using a 2.4GHz dual core CPU. Conclusion: A registration framework was developed to correct patient's setup according to register the planning CT volume data and the daily reconstructed 3D CBCT data. The experiments showed that the spatial MI algorithm can be adopted for head‐neck images. The improved Demons deformable registration was more suitable to lung images, and rigid alignment should be applied before deformable registration to get more accurate result. Supported by National Natural Science Foundation of China (NO.81101132) and Natural Science Foundation of Anhui Province (NO.11040606Q55)