應影像實驗室邀請,12月31日(周五)早上 10:00 –11:00,北京大學 姜明 教授給bet356手机版唯一官网師生做學術報告,具體情況如下:
報告題目:FPGA Acceleration for 3D Low-Dose CT
報告人:北京大學 姜明 教授
北京時間:2021年12月31日(周五)早上 10:00 –11:00
地點:線上報告
騰訊會議 鍊接:
會議主題:學術報告-北京大學 姜明 教授
會議時間:2021/12/31 10:00-11:00 (GMT+08:00) 中國标準時間 - 北京
點擊鍊接入會,或添加至會議列表:
https://meeting.tencent.com/dm/NoEjzXFHSpDl
#騰訊會議:658-402-560
摘要:
In this presentation, I am to talk about our recent progress on accelerating the image reconstruction for low dose X-ray CT with field-programmable gate array (FPGA) devices. Low-dose CT (LDCT) is to provide CT images of clinical quality with reduced cumulative radiation dose. Iterative image reconstruction methods with effective regularization are used for LDCT but generally induce higher computation load than the conventional filtered back-projection methods. The high computational demand of iterative image reconstruction methods with notably increased reconstruction time precludes its routine clinical application and makes it unacceptable in emergency. FPGA is highly programmable and enables optimizing the usage of computing resources at logic gate level. In our work, we formulate an iterative reconstruction algorithm by the Mumford-Shah regularization functional with beam-based asynchronous parallelism. In the implementation, because of the beam-based approach, we can use the pipelining and tiling techniques to increase the computational efficiency and reduce the memory I/O. The implementation is evaluated with a physical phantom and has image quality comparable with the vendor’s result. Relevant aspects of the asynchronous parallelism and implementation are also discussed. This is a joint work with Yong Cui, William Hsu, Guojie Luo, Linjun Qiao, and Wentai Zhang.
報告人簡介: