Head and neck synthetic-CT generation based on segmented B-spline deformable registration method(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2022年第1期
- Page:
- 44-50
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Head and neck synthetic-CT generation based on segmented B-spline deformable registration method
- Author(s):
- CHEN Lixia; QI Mengke; SONG Ting; LU Guangwen
- School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China
- Keywords:
- segmented B-spline deformable registration method head and neck synthetic-CT Dice similarity coefficient average absolute error
- PACS:
- R318
- DOI:
- DOI:10.3969/j.issn.1005-202X.2022.01.008
- Abstract:
- Abstract: Objective To explore the application of segmented B-spline deformable registration method in the generation of synthetic-CT (sCT) of the head and neck, and to discuss its effects on the accuracy of sCT generation. Methods A total of 45 cases of nasopharyngeal carcinoma treated with intensity-modulated radiotherapy were collected, each of which includes T1-weighted MRI images and CT images. 3D Slicer software was used to perform segmented B-spline deformable registration, global B-spline deformable registration, segmented rigid registration and global rigid registration for MRI and CT images. The Dice similarity coefficient (DSC) values of the MRI image after registration and real CT image were compared. Thirty of the patients were randomly selected as training set, and the other 15 cases as test set. The MRI and CT images after registration were trained to generate sCT through pix2pix network. The average absolute error (MAE), structural similarity coefficient (SSIM) and peak signal-to-noise ratio (PSNR) were compared between generated sCT and real CT. Moreover, the MAE of different tissues (bones, soft tissues, air and lipids) divided by threshold method was analyzed. Results Through the comparison on MRI after registration and real CT images, the DSC value of segmented B-spline deformable registration method was found to be optimal. The comparison on MAE, SSIM and PSNR between the real sCT and the sCT generated by 4 registration methods showed that the segmented registration method was better than the global registration method, and that B-spline deformable registration method was superior to rigid registration method. The MAE, SSIM and PSNR of segmented B-spline deformable registration method were (74.783±9.869) HU, 0.839±0.032 and (28.859±0.957) dB, respectively, which were better than those of the other registration methods. Segmented B-spline deformable registration method also achieved better MAE in the bones, soft tissues and lipids, but the MAE in the air was slightly worse than that of rigid registration method. Conclusion The segmented B-spline deformable registration method is superior to the global B-spline deformable registration method and rigid registration method in the head and neck sCT generation. The proposed method can improve the registration accuracy of head and neck images, thus enhancing the generation accuracy of sCT.
Last Update: 2022-01-17