Intrafractional rectum anatomy shape prediction based on 3D point cloud representation in online adaptive radiation therapy.
Authors
Affiliations (9)
Affiliations (9)
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- China North Artificial Intelligence & Innovation Research Institute, Beijing 100072, China; Collective Intelligence & Collaboration Laboratory, Beijing 100072, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China; Beijing Key Laboratory of Urologic Cancer Cell and Gene Therapy, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
- Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China. Electronic address: [email protected].
Abstract
This study aimed to develop an anatomical structure generative model to predict intrafractional rectal shape in prostate cancer online adaptive radiation therapy (OART). A retrospective analysis was conducted on clinical data from 42 prostate cancer patients treated with online adaptive radiotherapy (OART). Rectal shapes were extracted from MRI scans acquired at the pretreatment (Pre-) and position verification (Pv-) stages, and represented as 3D point clouds. Data augmentation was applied to construct the rectal dataset. Then, we developed SA-UNet, among the earliest generative AI-based models for intrafractional anatomical shape prediction, and benchmarked its performance against two conventional deep learning baseline models. The SA-UNet model demonstrated superior performance in anatomical structure prediction, yielding the lowest average CD (29.06 ± 12.56 mm) and EMD (4.82 ± 1.31 mm) values and the highest average JAC value (0.69 ± 0.07). Compared with Baseline-MLP, SA-UNet achieved significantly greater consistency across treatment fractions, with reduced variability and fewer outliers (p < 0.025, Bonferroni-adjusted). Meanwhile, the SA-UNet significantly outperformed Baseline-PointCNN, which had the highest average CD (42.85 ± 14.18 mm) and EMD (6.07 ± 1.31 mm), and the lowest JAC (0.62 ± 0.07), all significantly inferior to those of SA-UNet (p < 0.01, Bonferroni-adjusted). The SA-UNet model showed preliminary feasibility for intrafractional rectal shape prediction in OART, offering potential for early alerting and advancing precision radiotherapy.