四川大学学报（自然科学版）2025，Vol.62Issue(3)：556-568,13.DOI:10.19907/j.0490-6756.240373
基于生成负样本的对比约束无监督超声图像重建方法
Contrastively constrained unsuperivsed ultrasound reconstruction method based on negative sample generation
摘要
Abstract
Ultrasound plane-wave(PW)imaging achieves ultra-high frame rates but usually induces strong diffraction artifacts,which causes degradation of image quality.To improve the image quality,traditional re-construction approach consists of compounding successive transmissions coherently,at the expense of the frame rate.Deep learning-based ultrasound imaging has been proved to have the potential to improve the qual-ity of PW images.However,existing research predominantly focuses on supervised learning that relies on highly matched low-quality and high-quality paired inputs,which is difficult to achieve in real situations.Therefore,developing unsupervised methods that do not require paired inputs is more in line with practical needs.In this study,the authors introduce a novel unsupervised COntrastive-learning based model with Nega-tive sample Generation mechanism(CONG)for PW imaging.The framework integrates a generative learn-ing framework with contrastive learning and the corresponding negative sample generation mechanism,which effectively reduces diffraction artifacts in PW images and preserves structural details in reconstructed images so that improves the image quality.The authors evaluated the proposed approach with both numerical simula-tion and real acquisition.Experimental results demonstrate that the proposed approach achieve the best qualita-tive and quantitative performance compared with existing deep-learning based works.关键词
深度学习/GAN模型/超声图像重建/无监督图像重建/负样本生成Key words
Deep learning/GAN Models/Ultrasound image teconstruction/Unsupervised image reconstruc-tion/Negative sample generation分类
信息技术与安全科学引用本文复制引用
梁文卓,路景枫,王晖,张中洲,王志文,张意..基于生成负样本的对比约束无监督超声图像重建方法[J].四川大学学报（自然科学版）,2025,62(3):556-568,13.基金项目
国家自然科学基金(62401381) （62401381）
四川省博士后科研项目特别资助基金(TB2024074) （TB2024074）
