Computed tomography reconstruction using deep image prior and learned reconstruction methods
Реконструкция компьютерной томографии с использованием глубокого априорного представления изображения и обучаемых методов реконструкции
2020-07-08
SCID: 54.1/f4ygsvv5
Discuss with AI
computed tomography reconstructiondeep image priorlearned primal-dual methodlow-dose CTsparse-angle CT
Figures from the paper
Abstract (AI)
Abstract In this paper we describe an investigation into the application of deep learning methods for low-dose and sparse angle computed tomography using small training datasets. To motivate our work we review some of the existing approaches and obtain quantitative results after training them with different amounts of data. We find that the learned primal-dual method has an outstanding performance in terms of reconstruction quality and data efficiency. However, in general, end-to-end learned methods have two deficiencies: (a) a lack of classical guarantees in inverse problems and (b) the lack of generalization after training with insufficient data. To overcome these problems, we introduce the deep image prior approach in combination with classical regularization and an initial reconstruction. The proposed methods achieve the best results in the low-data regime in three challenging scenarios.
Key Findings
1
Combining deep image prior with classical regularization and an initial reconstruction addresses these limitations and achieves the best low-data results across three challenging scenarios.
2
End-to-end learned reconstruction methods lack classical inverse-problem guarantees and may fail to generalize when training data are insufficient.
3
The learned primal-dual method delivers outstanding reconstruction quality and data efficiency among evaluated learned approaches.
4
The study investigates deep-learning reconstruction for low-dose and sparse-angle computed tomography using small training datasets.
Research Object
low-dose and sparse-angle computed tomography reconstruction
Research Subject
reconstruction quality and data efficiency of deep-learning and deep-image-prior methods under limited training data, including generalization and inverse-problem guarantees
Publication Details
Publication Date
2020-07-08
Journal
Publisher
ISSN
Cited by
200
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest