Cross-domain data augmentation for deep-learning-based male pelvic organ segmentation in cone beam CT
(2020) Applied Sciences — (2020)
Files
Cross_domain_data_augmentation.pdf
Preprint
Open Access - Adobe PDF
- 1004.25 KB
Details
- Authors
Desbordes, Paul 
- Abstract
- For prostate cancer patients, large organ deformations occurring between radiotherapy treatment sessions create uncertainty about the doses delivered to the tumor and surrounding healthy organs. Segmenting those regions on cone beam CT (CBCT) scans acquired on treatment day would reduce such uncertainties. In this work,a 3D U-net deep-learning architecture was trained to segment the bladder, rectum, and prostate on CBCT scans. Due to the scarcity of contoured CBCT scans, the training set was augmented with CT scans already contoured in the current clinical workflow. Our network was then tested on 63 CBCT scans. The Dice similarity coefficient (DSC) increases significantly with the number of CBCT and CT scans in the training set,reaching 0.874, 0.814, and 0.758 for the bladder, rectum, and prostate respectively. This is about 10\% better than conventional approaches based on deformable image registration between planning CT and treatment CBCT scans, except for the prostate. Interestingly, adding 74 CT scans to the CBCT training set allowed to maintain high DSCs, while halving the number of CBCT scans. Hence, our work shows that although CBCT scans include artifacts, cross-domain augmentation of the training set is effective and can rely on large datasets available for planning CT scans.
- Affiliations
Citations
Léger, J., Brion, E., Desbordes, P., De Vleeschouwer, C., Lee, J., & Macq, B. (2020). Cross-domain data augmentation for deep-learning-based male pelvic organ segmentation in cone beam CT. Applied Sciences. Accepted/in-press. https://hdl.handle.net/2078.5/122370 (Original work published 2020)