Room: Karl Dean Ballroom B1
Purpose: In MRI-only proton radiotherapy workflow, a synthetic CT (SCT) is required for accurate dose calculation and planning. This study aims to develop a learning-based method to generate patient-specific SCT from routine anatomical MRI for MRI-only proton radiotherapy.
Methods: We proposed to incorporate patch-based anatomical signature into machine learning framework to iteratively predict CT from MRI based on auto-context model (ACM). Firstly, patch-based anatomical features were extracted from the aligned MRI-CT training images, and the most informative features were identified to train an ACM-based random forest. Then, we extracted the selected features from the new MRI and fed them into the well-trained forests for the CT prediction. Our predicted CT generated from MRI were registered to CT for generating a SCT-based proton treatment plan. Mean absolute error (MAE) and normalized cross-correlation (NCC) were used to quantify the differences between the SCT and planning CT. Clinically-relevant dose volume histogram (DVH) metrics were extracted from SCT-based and CT-based proton plans for quantitative dosimetric evaluation.
Results: This SCT prediction method was tested using 14 brain patients. The mean MAE between SCT and CT was 27.65Â±5.63HU and the mean NCC was 0.96Â±0.03 for all patients. The mean dose differences of the PTV, relevant organs-at-risk (OARs) on SCT and CT across all the plans were less than 1.0% for D10, D50, D95, Dmean and Dmax, respectively. No significant differences were observed in the DVH metrics of the PTV and OARs for both proton plans (p>0.05). The average pass rate of gamma analysis was over 99%.
Conclusion: We have developed a novel learning-based approach to generate MRI-based SCT for proton radiotherapy and demonstrated that proton therapy dose calculations based on the SCTs are in good agreement with those based on standard planning CT. Therefore, an MRI-only based proton treatment planning workflow is feasible for proton radiotherapy.
Not Applicable / None Entered.