摘要
由于低剂量CT情境下医学图像存在多样的噪声,其强度和种类各异,因此选择合适的算法对去噪至关重要。传统图像去噪方法基于先验知识,其优化过程相对繁琐,存在保留图像细节和处理效率方面的一定限制。相较之下,基于深度学习的去噪方法具备学习能力强大、非线性建模、端到端学习、适应性强和大规模并行计算等独特优势,使其相对于传统方法在处理复杂噪声场景时更为有效。本文全面概括并深入分析了当前低剂量CT图像去噪方法的研究热点。首先,简要介绍了低剂量CT图像去噪的步骤和过程。其次,结合当前基于深度学习的低剂量CT图像去噪方法的研究现状,重点探讨了残差学习、注意力网络以及自监督学习这三个最具代表性的研究热点,详细阐述了各种基础网络架构及其改进方法在低剂量CT图像去噪中的应用情况。最后,总结了当前低剂量CT图像降噪方法所面临的主要挑战,并提出了未来的研究方向,以促进低剂量CT图像去噪技术的进一步发展。
Due to the diverse nature of noise in medical images acquired through low-dose CT,the selection of appropriate algorithms for denoising is crucial.Traditional image denoising methods,relying on prior knowledge,require a relatively intricate optimization process,and have limitations in preserving image details and processing efficiency.In contrast,deep learning-based denoising methods exhibit unique advantages such as powerful learning capabilities,nonlinear modeling,end-to-end learning,strong adaptability,and efficient large-scale parallel compulation.Therefore,deep learning approaches are more effec-tive than traditional methods in complex noise scenarios.This paper comprehensively summarizes and delves into the current research focus of denoising methods for low-dose CT images.It begins with a brief overview of the processing in image denoising for low-dose CT.Subsequently,it explores the status of deep learning-based denoising methods for low-dose CT images with a focus on three representative research trends:residual learning,attention networks,and self-supervised learning.The study elaborates the application of various foundational network architectures and their modifications low-dose CT image denoising.Finally,the paper summarizes the main challenges faced by current low-dose CT image denoising methods and proposes future research directions to promote the further advancement of low-dose CT image denoising technology.
作者
蒲秋梅
沈林林
田景龙
韦洁瑶
PU Qiumei;SHEN Linlin;TIAN Jinglong;WEI Jieyao(School of Information Engineering,Minzu University of China,Beijing 100081,China;National Language Intelligent Analysis and Security Governance Key Laboratory of the Ministry of Education,Beijing 100081,China)
出处
《中国体视学与图像分析》
2023年第4期369-379,共11页
Chinese Journal of Stereology and Image Analysis
基金
国家自然科学基金资助项目(No.31971311)。
