Generation of Pseudo-CT using Two Sets of MRI Scans for MRI-only Radiation Therapy

Abstract

WWith increasing interests in magnetic resonance imaging (MRI)-only radiation therapy (RT) and the emergence of MRI system integrated with the linear accelerator, a method substituting the computed tomography (CT) in radiation therapy is required. The method to substitute the CT is by generating a pseudo-CT (pCT). A CT is necessary to provide the electron density information and to calculate the simulated dose distribution in treatment plans. In this thesis, a voxel-based method is developed to generate the pCT image using two sets of MRI data. The method is trained with the CT data and two different sets of MRI data of multiple patients, where the anatomical structures in the images are segmented into several regions. A regression analysis is used to determine the two-variable polynomial function for each region to relate a voxel's two MRI intensity values to its CT number. This method is validated by applying a leave-one-out-cross-validation (LOOCV) and the accuracy of the pCT is evaluated by determining the mean absolute error (MAE) comparing the pseudo-CT to the reference-CT. The average MAE across all patients is 40.3 ± 3.0 Hounsfield Unit (HU). Our proposed method shows promising results in using a multi-variable polynomial prediction model to predict CT numbers from MRI images. The generated pCT images closely match the reference-CT image and the MAE results are comparable to other studies using more complicated methods.