LKA-ADMM-CSNet在MRI高精度重建中的研究  

Research on LKA-ADMM-CSNet for high-precision MRI reconstruction

作　　者：王德成 于瓅[1] WANG Decheng;YU Li(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《哈尔滨商业大学学报(自然科学版)》2025年第1期59-66,共8页Journal of Harbin University of Commerce:Natural Sciences Edition

基　　金：2021安徽省重点研究与开发计划项目(No.202104d07020010)。

摘　　要：核磁共振图像的高精度重建研究可以更加准确地对患者进行诊断,能够及时地发现病患并进行针对性治疗,为临床决策提供了可靠的支撑.针对此影响,提出了一种改进的压缩感知算法,称为LKA-ADMM-CSNet算法,该算法通过将传统的模型压缩感知(CS)方法和数据驱动的深度学习结合起来,实现从稀疏采样中对测量图像进行重建.经过实验比较,对于快速CS复值核磁共振(MR)成像,与传统和其他深度学习方法相比,所提出的LKA-ADMM-CSNet相较于原有模型,其重构精度都得到了一定的提升,最好的重构精度提升了约30%,最差的重构精度则提升了6%左右,平均提升精度大约在16%.由此可见,提出的新模型在应用于MR成像上有着更优秀的表现.The high-precision reconstruction research of MRI images can provide more accurate diagnoses for patients,enabling timely detection and targeted treatment of diseases,thereby offering reliable support for clinical decision-making.In response to this impact,an improved compressed sensing algorithm was proposed,known as the LKA-ADMM-CSNet algorithm.This algorithm combines traditional model-based compressed sensing(CS)methods with data-driven deep learning to reconstruct measured images from sparse samples.Experimental comparisons showed that for fast CS complex-valued MRI imaging,compared to traditional methods and other deep learning approaches,the reconstructed precision of the proposed LKA-ADMM-CSNet model was significantly improved.The best reconstruction precision increased by about 30%,while the worst improved by approximately 6%,with an average improvement of around 16%.This demonstrated that the new model performed better in MRI imaging applications.

关 键 词：压缩感知 深度学习 MR成像 ADMM LKA LKA-ADMM-CSNet 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]