Sparse-view Cone Beam CT Reconstruction using Data-consistent Supervised and Adversarial Learning from Scarce Training Data
Abstract
Reconstruction of CT images from a limited set of projections through an object is important in several applications ranging from medical imaging to industrial settings. As the number of available projections decreases, traditional reconstruction techniques such as the Feldkamp, Davis, and Kress (FDK) algorithm and model based iterative reconstruction methods perform poorly.
Fig. 1: Flow diagram depicting the overall pipeline of our algorithm, where xFDK is the FDK reconstruction, xEP is an edge-preserving regularized reconstruction, xG is the generator’s output after slice aggregation, and x1 is the output of the first stage.
This work focuses on image reconstruction in such settings, namely, when both the number of available CT projections and the training data is extremely limited. We adopt a sequential reconstruction approach over several stages using an adversarially trained shallow network for "destreaking" followed by a data-consistency update in each stage. We use image subvolumes to train our method and patch aggregation and a hybrid 3D-to-2D mapping network for the "destreaking" component. Comparisons to other methods over several test examples indicate that the proposed method has much potential when both the number of projections and available training data are highly limited.
Results
Fig. 2: Comparison of the quality of reconstruction of our proposed algorithm (h) for walnut 2 (8 views) in Table I to various reference methods. Each
subfigure depicts slices through the center of the walnut volume in three different directions (or sagittal, coronal and transverse orientations). The normalized
mean absolute errors have also been shown underneath each subfigure. The central slices corresponding to the ground truth training walnut volume have also been shown in (b).
References
Anish Lahiri, Gabriel Maliakal, Marc L. Klasky, Jeffrey A. Fessler, and Saiprasad Ravishankar. "Sparse-view cone beam CT reconstruction using data-consistent supervised and adversarial learning from scarce training data." IEEE Transactions on Computational Imaging 9 (2023): 13-28. (Code)