Generation of Synthetic CBCT with Artifacts Caused by Peristaltic Movement From Planning CT
T Onishi1*, M Nakamura1, D Zhou1, Y Tsuruta1, M Sasaki1, N Matsushita1, A Ogawa1, R Ashida1, T Mizowaki1, (1) Kyoto University
Presentations
PO-GePV-I-11 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: To generate synthetic CBCT (sCBCT) with artifacts caused by peristaltic movement from planning CT (pCT). In the future, the paired pCT-sCBCT dataset can be used as training dataset for the deep learning model to reduce the artifacts.
Methods: Nine-hundred DRRs (1024 pixel x 768 pixel) were generated from pCT at 0.2-degree interval. Air regions in the intestines were segmented by HU values ranging from -1000 to -500 HU for generating air regions-only DRRs. To generate DRRs assuming to be acquired during intestine movement, morphological processing was conducted to repeatedly expand and contract the extracted air regions on the DRRs by changing the following parameters;mask_size (=3, 5, 7, 9 [pixel]): the size of masks to adjust the range over which the pixel values change uniformly. mask_number (=3, 5 ,7 ,11 ,15 ,19): the number of masks to adjust heights of the pixel value.By combining mask_size and mask_number, the gradient of the pixel value varies.grid_w (=2, 4, 8 [pixel]) and grid_h (=4, 8, 32, 256 [pixel]): grid width and height dividing DRRs for adaptive adjustment of pixel values, which adjusted the smooth transition of pixel values.sCBCT was reconstructed with the Feldkamp-Davis-Kress algorithm. Visual similarity of sCBCT to clinical CBCT was rated for a total of 288 sCBCT images, by 2 radiation oncologists, 3 radiotherapy technologists and 1 medical physicist in a scale from 1 to 3 (with 1 being least similarity and 3 being most similarity).
Results: The percentage that average scores were higher than 2 were 46% for mask_num of 7 and 11, 30% with mask_size for 3, 36% for grid_w of 2, and 23% for grid_h of 32.
Conclusion: sCBCT with artifacts caused by peristaltic movement was successfully generated from pCT. Our technique can be used to generate training datasets for the paired pCT-sCBCT dataset.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email