Title:
SELF-SUPERVISED LEARNING METHOD AND APPLICATION
Document Type and Number:
WIPO Patent Application WO/2022/011690
Kind Code:
A1
Abstract:
A self-supervised learning method and an application, the method comprising: modeling image noise; generating an approximate target pixel value of a current pixel from a low-dose CT image, and obtaining a target pixel point; randomly cropping image blocks from a low-dose CT input image, randomly selecting N pixel points from the image blocks, and replacing the target pixel points with currently selected pixel points to obtain a target image; and training a network to gradually reach a state of convergence. Thus, a network can be trained end-to-end without manual intervention, and noise reduction can be achieved.
More Like This:
Inventors:
JIANG HONGWEI (CN)
ZHENG HAIRONG (CN)
LI YANMING (CN)
WAN LIWEN (CN)
ZHENG HAIRONG (CN)
LI YANMING (CN)
WAN LIWEN (CN)
Application Number:
PCT/CN2020/102732
Publication Date:
January 20, 2022
Filing Date:
July 17, 2020
Export Citation:
Assignee:
NAT INSTITUTE OF ADVANCED MEDICAL DEVICES SHENZHEN (CN)
International Classes:
G06N3/00
Domestic Patent References:
WO2019147767A1 | 2019-08-01 |
Foreign References:
CN111260055A | 2020-06-09 | |||
CN109035169A | 2018-12-18 | |||
CN110599420A | 2019-12-20 |
Attorney, Agent or Firm:
BEIJING CHENGHUI LAW FIRM (CN)
Download PDF: