School of Physics, Beihang University Beijing, Beijing
T. Lv1, Y. Zhang2, Y. Wang3, X. Chen3, K. Men3, and W. Zhao1,4; 1School of physics, Beihang University, Beijing, China, 2Shanghai General Hospital, Shanghai, China, 3National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, 4Beihang Hangzhou Innovation Institute, Hangzhou, Zhejiang, China
Purpose/Objective(s): Patients treated with left-sided breast radiotherapy may benefit from the application of the deep inspiration breath-hold (DIBH) or the free-breathing (FB) technique. Both FB-CT and DIBH-CT scans are required to make decisions regarding the radiotherapy modality. However, rescanning of DIBH-CT increases the received dose and treatment cost. We propose a patient-specific based strategy for predicting the DIBH (pDIBH) CT from FB-CT and DIBH orthogonal projections. Materials/
Methods: A total of 185 breast cancer patients underwent scans with both FB-CT and DIBH-CT, with 144 for training and 41 for testing. A synthetic network was trained to predict DIBH-CT from FB-CT. For each patient in the test set, the model was re-trained using FB-CT and DIBH orthogonal projections to obtain the final synthetic DIBH-CT (sDIBH-CT). A combined Demons image registration and material decomposition algorithm was utilized to register FB-CT to sDIBH-CT to obtain the pDIBH-CT and deformation vector field (DVF). Since the mean HU values in the lungs for FB-CT and DIBH-CT were different, the HU values in the lungs for pDIBH-CT were adjusted using a histogram matching algorithm. The DVF were geometrically assessed using the Dice similarity coefficient (DSC), and image quality of pDIBH-CT were assessed using mean absolute errors (MAE). Results: The DSC between DIBH-CT, FB-CT, and pDIBH-CT were improved from 0.778±0.052 to 0.960±0.019 for lung L,0.786±0.050 to 0.965±0.010 for lung R, and 0.743±0.066 to 0.902±0.028 for heart, respectively. The MAE were decreased from 185.7±39.1 HU to 55.2±9.7 HU for body,362.5±57.5 HU to 85.2±7.6 HU for lungs, and 52.1±24.8 HU to 22.0±5.2 HU for heart, respectively. The pDIBH-CT was comparable to DIBH-CT in terms of geometry and image quality. Conclusion: The proposed strategy can predict the DIBH-CT from FB-CT and DIBH orthogonal projections. The pDIBH-CT shows great consistency with scanned DIBH-CT in both geometric and image quality dimensions. This strategy avoids the need for patients to rescan DIBH-CT, reducing the treatment cost and dose received. The pDIBH-CT has the potential to replace DIBH-CT in left-sided breast radiotherapy modality decisions.
