TransCBCT: Improving the Image Quality of Cone-Beam Computed Tomography with Transformer
X Chen1*, Y Liu1, B Yang1, J Zhu1, Y Liu1, J Dai1, K Men1, (1) National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences, Beijing, 100021,CN
Presentations
PO-GePV-M-320 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: The challenge of cone-beam computed tomography (CBCT) is its low image quality, which limits its application for adaptive radiation therapy (ART). Transformers have recently received significant success in computer vision, which employs multi-head attention mechanisms to capture long-range contextual relations between image pixels. Spurred by the advantages of transformers, we proposed a novel transformer-based network (called TransCBCT) to improve the image quality of CBCT.This study aimed to demonstrate that (i) transformer can be adopted in image improvement for ART application, which needs to preserve the structure of raw CBCT and calibrate the CT numbers. (ii) Importing the transformer block can be more powerful than a conventional pure convolution-based network.
Methods: In this study, 91 patients diagnosed with prostate cancer were enrolled. We constructed a transformer-based hierarchical encoder–decoder structure with skip connection, called TransCBCT. The network also employed several convolutional layers to capture local context. The proposed TransCBCT was trained on 6,144 paired CBCT/deformed CT images from 76 patients and tested on 1,026 paired images from 15 patients. The performance of the proposed TransCBCT was compared with a widely recognized style transferring deep learning method, the cycle-consistent adversarial network (CycleGAN).
Results: TransCBCT had superior performance in improving the image quality of CBCT. The mean square error of TransCBCT was 28.8 ± 16.7 HU, compared to 66.5 ± 13.2 for raw CBCT and 34.3 ± 17.3 for CycleGAN. It can preserve the structure of raw CBCT and reduce artifacts and noises. The proposed TransCBCT also achieved promising results for calibrating the CT numbers, which was better than CycleGAN.
Conclusion: The proposed TransCBCT can effectively improve the image quality of CBCT. It has immense potential to improve radiotherapy accuracy.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the National Natural Science Foundation of China (12175312, 11975313, 12005302), Beijing Nova Program (Z201100006820058), and the CAMS Innovation Fund for Medical Sciences (2020-I2M-C&T-B-073, 2021-I2M-C&T-A-016).
Keywords
Cone-beam CT, Computer Vision, Image Artifacts
Taxonomy
IM- Cone Beam CT: Machine learning, computer vision
Contact Email