Author: Sean L. Berry, Weixing Cai, Laura I. Cervino, Yabo Fu, Daphna Gelblum, Wendy B. Harris, Xiuxiu He, Licheng Kuo, Tianfang Li, Xiang Li, Jean M. Moran, Boris Mueller, Huiqiao Xie, Mitchell Yu, Hao Zhang 👨🔬
Affiliation: Department of Radiation Oncology, Memorial Sloan Kettering Cancer Center, Department of Medical Physics, Memorial Sloan Kettering Cancer Center 🌍
Purpose: Gating ablative radiotherapy for pancreatic cancer accounts for tumor movement due to respiration and typically requires 5, 15, or 25 fractions. Pretreatment imaging verification is essential for safe delivery, but conventional gated CBCT on C-arm linear accelerators takes 2-8 minutes per scan. This study introduces an innovative nonstop gated CBCT imaging paradigm that reduces the scan time to just 1 minute and decreases imaging dose by over 40%. A deep learning-based dual-domain reconstruction network is proposed to achieve high reconstruction efficiency and superior image quality.
Methods: One-minute nonstop gated CBCT imaging is achieved by continuous gantry rotation while the kV beam is triggered according to respiratory gating signals, resulting in under-sampled and non-uniform projection data. Nonstop gated CBCT scans were emulated by down-sampling raw gated CBCT projection data of gating ablative pancreatic patients based on their respiratory gating signals. A dual-domain convolutional neural network (DDCNN) was developed to address the challenges of high-quality image reconstruction using both projection-domain and image-domain CNNs. The DDCNN results, utilizing 55 scans from 12 patients, were compared with current clinically available reconstruction methods, including the FDK algorithm and the penalized-likelihood iterative method (PL).
Results: Both the FDK and PL methods struggled to reconstruct the under-sampled and non-uniform projection data from nonstop gated CBCT acquisitions. In contrast, the DDCNN approach significantly improved image quality, enhanced fiducial marker visibility, and reduced artifacts. Although soft tissue contrast is generally limited in abdominal CBCT scans, the DDCNN method recovered more contrast details. Furthermore, DDCNN reconstruction was faster than iterative methods, offering substantial time efficiency.
Conclusion: A time- and dose-efficient nonstop gated CBCT imaging technique, coupled with a DDCNN reconstruction framework, has been proposed and validated. This innovative approach has the potential to substantially reduce patient on-table time, alleviating patient discomfort and anxiety during gating pancreatic ablative radiotherapy.