Deep learning-based relative stopping power mapping generation with cone-beam CT in proton radiation therapy

Abstract

Purpose:Dual-energy CT (DECT) has been used to derive relative stopping power (RSP) map by obtaining the energy dependence of photon interactions. The DECT-derived RSP maps could potentially be compromised by image noise levels and the severity of artifacts when using physics-based mapping techniques, which would affect subsequent clinical applications. This work presents a noise-robust learning-based method to predict RSP maps from DECT for proton radiation therapy.Method:The proposed method uses a residual attention cycle-consistent generative adversarial network. Residual blocks with attention gates were used to force the model focus on the difference between RSP maps and DECT images. Cycle consistent generative adversarial networks were used to let the DECT-to-RSP mapping be close to a one-to-one mapping by introducing an inverse RSP-to-DECT mapping. To evaluate the accuracy of the method, we retrospectively investigated 20 head-and-neck cancer patients with DECT scans in proton radiation therapy simulation. Ground truth RSP values were assigned by calculation based on chemical compositions. These ground truth RSP maps acted as learning targets in the training process for DECT datasets, and were evaluated against results from the proposed method using a leave-one-out cross-validation strategy.Result:The predicted RSP maps showed an average normalized mean square error (NMSE) of 2.83% across the whole body volume, and average mean error (ME) less than 3% in all volumes of interest (VOIs). With additional simulated noise added in DECT datasets, the proposed method still maintained a comparable performance, while the physics-based stoichiometric method suffered degraded inaccuracy from increased noise level. Based on the statistical analysis of comparative dose volume histogram (DVH) metrics of dose maps calculated on ground truth and predicted RSP maps among 19 pencil beam scanning proton treatment plans, the average differences in DVH metrics for clinical target volumes (CTVs) were less than 0.2 Gy for D95% and Dmax with no statistical significance (average prescription dose = 60 Gy). Maximum difference in DVH metrics of organs-at-risk (OARs) was around 1 Gy on average. Conclusion:These results strongly indicate the high accuracy of RSP maps predicted by our machinelearning-based method and show its potential feasibility for proton treatment planning and dose calculation.