Truncation occurs due to limited detector size. The SOTA method for truncation correction is a plug-and-play method proposed by Huang [10.1109/TMI.2021.3072568], in which the generative model like Pix2pixGan is used. Recently the score-based generative model has achieved better results compared to GAN and been applied in medical image processing tasks like sparse-view CT reconstruction [https://arxiv.org/abs/2111.08005] and MRI [https://doi.org/10.1016/j.media.2022.102479]. Such models can be applied in the truncation correction task as well and the prior information from truncated images could further improve the performance.
Requirement: 1) Knowledge of CT reconstruction algorithm. 2) Experience in Deep Learning.