Program Information
Evaluation of Pelvic Synthetic CT Images Generated by the Statistical Decomposition Algorithm Using Small Field-Of-View T2-Weighted MRI Images
J Fu1*, K Singhrao1 , D Ruan2 , Y Yang2 , J Lewis2 , (1) University of California Los Angeles, Los Angeles, CA, (2) UCLA School of Medicine, Los Angeles, CA
Presentations
MO-L-GePD-J(B)-6 (Monday, July 31, 2017) 1:15 PM - 1:45 PM Room: Joint Imaging-Therapy ePoster Lounge - B
Purpose: To quantify the accuracy of pelvic synthetic CT images generated by the Statistical Decomposition Algorithm (SDA). MRI-only radiotherapy has drawn growing research interest due to limitations in using CT simulation in current MRI-guided radiotherapy workflows. These limitations include systematic registration errors and additional costs. In order to use MRI as the sole modality for treatment simulation, the electron density map must be acquired directly and precisely from MRI scans.
Methods: The data consists of CT scans and small field-of-view (FOV) T2-weighted MRI scans of nine prostate cancer patients. The SDA, integrated in the commercial software MriPlanner, was used to generate synthetic CT images for each patient. For comparison, real CT images were rigidly registered to synthetic CT images. The results were evaluated using the voxel-wise mean absolute error (MAE) and the mean error (ME) of CT numbers, and the Dice similarity coefficient (DSC) of bone volume to measure geometric consistency.
Results: The MAE and ME of CT numbers between real and synthetic CT images across nine patients are 62.4±12.0 HU and -26.2±15.0 HU, respectively. The MAE curve shows there is the largest error in high-density bone and air regions. Mean DSC of bone volume is 0.77±0.04 when bone is identified by thresholding at 150 HU. As the threshold increases, corresponding to regions of higher bone density, DSC gradually decreases. The SDA was not able to generate gas cavities, or calcifications in Synthetic CT images. In some cases, vertebrae were missing, possibly because of the small FOV.
Conclusion: The SDA provides reasonable predictions for CT number using only T2-weighted MRI. However, our preliminary small FOV results revealed some discrepancies for structures with high CT number. The dosimetric accuracy will be examined in a future study.
Contact Email: