2021 Fiscal Year Research-status Report
Large-scale Tomography Computation
| Project/Area Number |
21K17750
|
|---|
| Research Institution | National Institute of Advanced Industrial Science and Technology |
|---|
Principal Investigator |
陳 鵬 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 研究員 (30890199)
|
|---|
| Project Period (FY) |
2021-04-01 – 2025-03-31
|
| Keywords | Computed Tomography / Image Reconstruction / Filtered Back-projection / CPU / GPU |
|---|
| Outline of Annual Research Achievements |
We propose a novel decomposition scheme and reconstruction algorithm for distributed FPB algorithm. This scheme enables arbitrarily large input/output sizes, eliminates the redundancy arising in the end-to-end pipeline, and improves the scalability by replacing two communication collectives with only one segmented reduction. We implement the proposed decomposition scheme for input and output problems in a framework that is useful for all current-generation CT devices (7th gen). In our experiments using up to 1024 GPUs, our framework can construct 4096x4096x4096 volumes, for real-world datasets, in under 16 seconds.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
The back-projection operation is a fundamental computing kernel for image reconstructions, e.g. Filtered Back-projection (FBP). We propose a collection of novel back-projection algorithms that reduce the arithmetic computation, robustly enable vectorization, enforce a regular memory access pattern, and maximize the data locality. We also implement the novel algorithms as efficient back-projection kernels that are performance portable over a wide range of CPUs.
We propose a novel decomposition scheme and reconstruction algorithm for distributed FPB algorithm. This scheme enables arbitrarily large input/output sizes, eliminates the redundancy arising in the end-to-end pipeline and improves the scalability by replacing two communication collectives with only one segmented reduction.
|
| Strategy for Future Research Activity |
We plan to optimize the computing kernels using heterogeneous computing architectures for achieving computing efficiency and power efficiency, e.g., research on the FPGA accelerator. Also, we intend to extend our framework for iterative image reconstruction algorithms.
|
| Causes of Carryover |
- Personal: The research personnel will be hired to carry out the research plan. The roles would include solving research problems as well as engineering and development.
- Travel and publications: The travel and publications costs are crucial for disseminating the work in this project. The costs would cover the attendance of conferences and workshops, inside and outside Japan to receive feedback at the early stages and publicize the research findings at the later stages.
- Equipment: Research participants will use personal PCs for conducting basic research workflow.
|
-
-
-
-
-
[Journal Article] 特集 映像情報処理のための高性能計算基盤2021
Author(s)
Chen Peng、Mohamed Wahib, Yusuke Tanimura, Hirotaka Ogawa, Satoshi Matsuoka
-
Journal Title
映像情報メディア学会誌
Volume: 75 (5)
Pages: 597-602
Peer Reviewed / Open Access / Int'l Joint Research
-
-
-
[Presentation] Scalable FBP decomposition for cone-beam CT reconstruction2021
Author(s)
Chen Peng、Wahib Mohamed、Wang Xiao、Hirofuchi Takahiro、Ogawa Hirotaka、Biguri Ander、Boardman Richard、Blumensath Thomas、Matsuoka Satoshi
Organizer
SC '21: Proceedings of the International Conference for High Performance Computing, Networking, Storage and AnalysisNovember 2021
Int'l Joint Research / Invited
-
-
