Room: Exhibit Hall | Forum 2
Purpose: Increasing interests in using magnetic resonance imaging (MRI) only in radiotherapy treatment require methods for predicting the computed tomography (CT) number and generating the pseudo-CT (pCT) image from MRI. This study aims to develop a voxel-based method to generate pseudo-CT images from two sets of MRI data for dose calculations in radiotherapy treatment plans.
Methods: Pelvic MRI (T1-Flash, T2-TSER) and CT scan data are collected from six patients with carcinoma of the cervix in a prospective study. The CT data and the two different sets of MRI data of multiple patients are segmented into several regions. A regression analysis is used to determine a two-variable polynomial function for each region to relate voxelâ€™s two MRI intensity values to their average CT number. The method is validated by applying a leave-one-out-cross-validation (LOOCV), where the model is trained with data from five patients and then applied to the MRI data of the remaining patient to generate the pseudo-CT. We evaluate the accuracy via mean absolute error (MAE) to find the Hounsfield unit (HU) difference between the pCT and reference-CT (rCT) images.
Results: The mean absolute error across all patients is 40.3 Â± 3.0 HU, ranging from 36.7 to 44.4 HU for individual LOOCV cycle.
Conclusion: The proposed voxel-based method generates pCT images with MAE results in close agreement with rCT images. Our MAE results are better than previous results from voxel-based methods or most atlas-based methods and comparable to another study using the more complicated atlas-based method.