降噪
计算机科学
噪音(视频)
人工智能
缩小
主成分分析
压缩传感
迭代重建
模式识别(心理学)
算法
可靠性(半导体)
信号(编程语言)
信噪比(成像)
噪声测量
稳健性(进化)
计算机视觉
数据采集
磁共振成像
信号处理
合成数据
规范(哲学)
医学影像学
灵敏度(控制系统)
图像噪声
散粒噪声
还原(数学)
估计理论
特征提取
作者
Chenlu Guo,Caohui Duan,Feiyang Lou,Armin Michael Nagel,Xin Lou,Xiaotong Zhang
标识
DOI:10.1109/tim.2026.3654722
摘要
Quantitative sodium (²³Na) magnetic resonance imaging (MRI) provides unique metabolic information but is hampered by inherently low signal-to-noise ratio (SNR), which compromises the accuracy and precision of tissue sodium concentration (TSC) measurements. To address this fundamental measurement challenge, we propose a novel reconstruction framework designed to enhance the reliability of TSC quantification. Our method, termed MC-WSNM-GRPCA, models the multi-dimensional ²³Na MRI data by integrating multi-channel weighted nuclear norm minimization with generalized robust principal component analysis. This model-based denoising approach leverages inter-slice noise variations and flexible low-rank approximations to robustly separate the low-energy ²³Na signal from noise components in the image domain. We evaluated the framework using simulated data corrupted by various noise types, including Gaussian, Rician, and acquisition noise, and in vivo data from glioma patients, acquired with a 32-channel ²³Na/¹H head coil. Compared to several state-of-the-art methods, our approach demonstrated superior performance in both noise suppression and preservation of underlying signal fidelity, as evaluated by quantitative metrics and visual quality. In conclusion, the proposed model-based reconstruction framework effectively improves the precision of TSC quantification from low-SNR ²³Na MRI data. This work provides a promising solution for enhancing the clinical value of quantitative sodium imaging in applications such as early diagnosis of brain tumors and stroke.
科研通智能强力驱动
Strongly Powered by AbleSci AI