Room: Track 1
Purpose: Dual-energy CT (DECT) has been shown to derive stopping power ratio (SPR) maps with higher accuracy than conventional single energy CT (SECT) by obtaining the energy dependence of photon interactions. However, DECT is not as widely implemented as SECT in proton radiation therapy (PRT) simulation due to the costs associated with acquiring and operating DECT. This work presents a learning-based method to synthesize DECT images from SECT for PRT.
Methods: The proposed method uses a residual generative adversarial network. Residual blocks with attention gates were used to force the model focus on the difference between DECT and SECT images. To evaluate the accuracy of the method, we retrospectively investigated 20 head-and-neck cancer patients with both DECT and SECT scans available. The high and low energy CT images acquired from DECT acted as learning targets for SECT datasets and were evaluated against results from the proposed method using a leave-one-out cross-validation strategy. To quantify the prediction quality of synthetic DECT images, three commonly used metrics were applied, including mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and normalized cross correlation (NCC).
Results: The mean ± standard deviation (SD) of MAE, PSNR and NCC for synthetic high energy CT image were 30.23 ± 4.02 HU, 31.83 ± 0.98 dB, and 0.97 ± 0.01, respectively. The mean±SD of MAE, PSNR and NCC for synthetic low energy CT image were 28.57 ± 5.03 HU, 32.02 ± 1.11 dB, and 0.97 ± 0.01, respectively. The corresponding SPR maps generated from synthetic DECT showed an average normalized mean square error of about 1% and have significantly reduced noise level and artifacts than those from original DECT.
Conclusion: These results strongly indicate the high accuracy of synthesized DECT images by our deep-learning-based method and show its potential feasibility for proton treatment planning and dose calculation.