Due to its wide field of view, cone-beam computed tomography (CBCT) is plagued by large amounts of scatter, where attenuated photons hit the detector, and corrupt the linear models used for reconstruction. Given that one can generate a good estimate of scatter however, then image accuracy can be retained. In the context of adaptive radiotherapy, one usually has a low-scatter planning CT image of the same patient at an earlier time. Correcting for scatter in the subsequent CBCT scan can either be self consistent with the new measurements or exploit the prior image, and there are several recent methods that report high accuracy with the latter. In this study, we will look at the accuracy of various scatter estimation methods, how they can be effectively incorporated into a statistical reconstruction algorithm, along with introducing a method for matching off-line Monte-Carlo (MC) prior estimates to the new measurements. Conclusions we draw from testing on a neck cancer patient are: statistical reconstruction that incorporates the scatter estimate significantly outperforms analytic and iterative methods with pre-correction; and although the most accurate scatter estimates can be made from the MC on planning image, they only offer a slight advantage over the measurement based scatter kernel superposition (SKS) in reconstruction error.
|Communications in Computer and Information Science
|21st Annual Conference on Medical Image Understanding and Analysis, MIUA 2017
|11/07/17 → 13/07/17
- Computed tomography
- Scatter estimation
- Prior information