Room: Karl Dean Ballroom B1
Purpose: To develop a generative adversarial network (GAN) method to produce synthetic CT images with accurate geometric and electron density information from MRI.
Methods: A deep GAN model was designed to learn a direct mapping function to convert a MRI slice to its corresponding CT slice. The model was trained by collecting all MRI slices with corresponding CT slices from each training subjectâ€™s MRI/CT pair. One hundred brain tumor patients with both CT and T2-weighted MRI images was used for this study.
Results: The proposed GAN method produced an overall average structural similarity index (SSIM) of 84 Â± 0.06 HU for all subjects. The overall mean square error MSE was 0.006 Â± 0.005 for all subjects. Although training a GAN model can be slow, training only needs to be done once. Applying a trained model to generate a complete synthetic CT volume for each new patient MRI image took only 9 s, which is much faster than an atlas-based approach.
Conclusion: A GAN model method was developed, and shown to be able to produce highly accurate synthetic CT estimations from conventional, single-sequence MRI images in near real time. Quantitative results also showed that the proposed method can generate synthetic CT images with improved accuracy and faster computation speed compared to conventional methods. Further validation of dose computation accuracy and on a larger patient cohort is warranted. Extensions of this method are also possible to further improve accuracy or handle multi-sequence MRI images.