Room: Karl Dean Ballroom B1
Purpose: To develop a learning-based method to generate patient-specific synthetic CT (SCT) from routine anatomical MRI for MRI-only prostate radiotherapy.
Methods: We proposed to integrate auto-context model (ACM) and patch-based anatomical signature into machine learning framework to effectively capture the relationship between the prostate CT and MRI for CT synthesis. The proposed prediction of SCT consists of two major stages: the training and the prediction stages. During the training stage, the 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. During the prediction stage, we extracted the selected features from the new MRI and fed them into the well-trained forests for the CT prediction. The predicted CT generated from MRI were registered to CT to generate a SCT-based treatment plan. Mean absolute error (MAE) and normalized cross-correlation (NCC) were used to quantify the differences between the SCT and CT. Clinically-relevant dose volume histogram (DVH) metrics were extracted from the SCT-based and CT-based plans for quantitative dosimetric evaluation.
Results: This CT prediction method was clinically tested with 34 treatment plans from 22 prostate patients. The mean MAE and NCC between SCT and CT were 29.86Â±10.04HU and 0.98Â±0.03 for all patients. Mean dose differences of the PTV, bladder, rectum and femur head on SCT and CT across all the plans were less than 0.50% for Dmin, D10, D50, Dmean and Dmax, respectively. No significant differences were observed in the DVH metrics of the PTV and organs-at-risk for both plans (p>0.05).
Conclusion: We have developed a novel learning-based approach to generate prostate MRI-based SCT and demonstrated its feasibility and reliability. The image similarity between SCT and CT and dosimetric agreement between SCT-based and CT-based dose plans warrants further study and development of a MRI-only workflow for prostate radiotherapy.
Not Applicable / None Entered.