Title:
AUTOMATIC DATA AUGMENTATION-BASED MEDICAL IMAGE SEGMENTATION METHOD
Document Type and Number:
WIPO Patent Application WO/2023/197612
Kind Code:
A1
Abstract:
Disclosed is an automatic data augmentation-based medical image segmentation method, comprising: S1, randomly dividing an original training set into a training set and a verification set according to a set proportion; S2, building a data augmentation search space to obtain a sampling sub-strategy; S3, training the sampling sub-strategy on the training set, in each iteration, updating a network weight by means of stochastic gradient descent, using the updated network weight to calculate verification set loss, updating a strategy parameter by means of proximal iteration, and when the verification set loss is the minimum, obtaining data augmentation sub-strategies; and S4, in a re-training stage, applying the data augmentation sub-strategy to the original training set for data augmentation, performing training on the augmented training set to obtain an optimal network weight, and using the network weight for reasoning to obtain a target area. The algorithm used in the present invention achieves advanced performance under a basic network architecture, and the efficiency of the search strategy of the algorithm is at least improved by one order of magnitude.
Inventors:
LIU MIN (CN)
LIU QINGHAO (CN)
ZHANG ZHE (CN)
FAN WENPEI (CN)
WANG YAONAN (CN)
LIU QINGHAO (CN)
ZHANG ZHE (CN)
FAN WENPEI (CN)
WANG YAONAN (CN)
Application Number:
PCT/CN2022/134722
Publication Date:
October 19, 2023
Filing Date:
November 28, 2022
Export Citation:
Assignee:
UNIV HUNAN (CN)
International Classes:
G06T5/00
Foreign References:
| CN114693935A | 2022-07-01 | |||
| CN112686282A | 2021-04-20 | |||
| CN111882492A | 2020-11-03 | |||
| CN112651892A | 2021-04-13 | |||
| KR20210033235A | 2021-03-26 |
Other References:
QUANMING YAO; JU XU; WEI-WEI TU; ZHANXING ZHU: "Efficient Neural Architecture Search via Proximal Iterations", ARXIV.ORG, 30 May 2019 (2019-05-30), XP081536513
Attorney, Agent or Firm:
BEIJING WEISHENGDA INTELLECTUAL PROPERTY AGENCY (GENERAL PARTNERSHIP) (CN)
Download PDF: