首页|期刊导航|实验技术与管理|X射线CT图像处理方法与定量分析研究进展

X射线CT图像处理方法与定量分析研究进展OA

Advances in X-ray computed tomography image processing methods and quantitative analysis

中文摘要英文摘要

X 射线计算机断层成像(X-CT)是一种重要的无损成像技术,用于多孔介质内部结构表征,在地理学、材料科学、生物医学等学科有重要应用.X-CT 图像为多角度投影经数学重建获得的三维衰减场,灰度值与物质类别并非简单对应,数据处理流程直接影响定量分析的可靠性.噪声、伪影、分辨率限制与灰度重叠等是制约 X-CT 定量精度的主要因素,其影响沿处理流程链传递并放大,导致孔隙率、连通性及拓扑参数等指标偏差.围绕 X-CT 图像处理全链条,文章系统梳理了图像重建、预处理、分割、三维表征及参数提取等环节的主流方法与适用局限,并讨论了 AI 辅助分析、三维多模态融合及标准化可复现分析等前沿方向.未来,X-CT 图像处理需在算法稳健性、流程透明性与结果可验证性间寻求平衡,以提升复杂样品定量分析的可信度与跨研究可比性.

[Objective]X-ray computed tomography(X-CT)is an important nondestructive three-dimensional imaging technique for characterizing internal structures across various fields,including geography,geoscience,materials science,biomedicine,and porous media research.Compared with destructive testing and two-dimensional section observation,X-CT enables the digital reconstruction of internal structures,providing a basis for quantitative characterization,geometric measurement,and process-oriented analysis.However,CT images are not direct optical records of a sample's true structure;rather,they represent three-dimensional attenuation fields reconstructed from multi-angle projection data through mathematical algorithms.The grayscale value of each voxel depends on multiple factors,including material composition,density,X-ray energy spectrum,imaging geometry,detector response,sampling strategy,and reconstruction method.Therefore,grayscale values lack a simple one-to-one correspondence with material categories.Reliable structural interpretation requires a comprehensive processing workflow that transforms raw grayscale volume data into physically meaningful,computable,and comparable quantitative information.[Methods]Exploring the full X-CT processing chain,this review systematically summarizes the major methods,technical roles,and applicable limitations of image reconstruction,preprocessing,artifact correction,target segmentation,three-dimensional visualization,structural quantification,parameter extraction,and common software platforms.Image reconstruction determines the quality of the input data for subsequent processing,while preprocessing and artifact correction improve the signal-to-noise ratio,reduce systematic distortions,and enhance structural boundaries.Segmentation is the key step that converts continuous grayscale data into discrete structural categories such as pores,particles,fractures,matrix phases,or defects.Three-dimensional characterization and parameter extraction further generate quantitative indicators,including porosity,specific surface area,connectivity,tortuosity,and pore-network topology.Because these parameters derive from a chain of interdependent procedures,the reliability of CT-based quantitative analysis depends not only on individual algorithms but also on the consistency,transparency,and quality control of the entire workflow.[Results]The review identifies noise,artifacts,insufficient spatial resolution,partial-volume effects,and grayscale overlap as the major factors restricting the accuracy and robustness of CT image analysis.Noise and artifacts may distort grayscale distributions and structural boundaries;insufficient spatial resolution and partial-volume effects may blur interfaces and lead to the omission or underestimation of small-scale pores,throats,and fractures;grayscale overlap in complex multiphase samples often triggers threshold-based segmentation instability.These uncertainties propagate and amplify along the processing chain,impacting the accuracy and comparability of porosity,connectivity,specific surface area,tortuosity,and topological parameters.In addition,differences in preprocessing strategies,segmentation algorithms,region-of-interest selection,parameter definitions,and software implementations may reduce reproducibility across studies.Therefore,CT-derived parameters must be understood as structural estimates under specific imaging conditions and processing workflows rather than absolute representations of the true structure.[Conclusions]Recent advances in artificial intelligence,three-dimensional quantitative analysis,multimodal fusion,and reproducible workflows are transforming X-CT image processing.Machine learning and deep learning methods provide new opportunities for denoising,artifact correction,and complex-structure segmentation,especially in low-contrast,noisy,or large-scale three-dimensional datasets.Nevertheless,model performance depends strongly on training-data representativeness,annotation quality,imaging conditions,generalization,and physical interpretability.By pairing X-CT with complementary techniques,such as FIB-SEM,EDS,EBSD,NMRI,or spectral imaging,multimodal fusion integrates structural,compositional,and functional data across scales,thereby extending CT analysis from morphological description to structure-composition-function interpretation.At the same time,standardized reporting,open data,open-source workflows,benchmark datasets,and uncertainty assessment are essential for improving reproducibility and cross-study comparability.Future X-CT image processing should not simply pursue algorithmic complexity or full automation.Instead,it must balance algorithm robustness,workflow transparency,and result verifiability,thereby enhancing the credibility of quantitative analysis of complex samples and strengthening the role of X-CT in cross-scale structural characterization and multidisciplinary research.

胡广荣;龚吉蕊;刘莹莹;高晓飞;刘春玲;魏欣;刘瑛娜;辛甜

北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875北京师范大学 地理科学学部 地表过程与水土风沙灾害风险防控全国重点实验室,北京 100875

社会科学

X射线CT图像处理图像分割三维结构定量分析伪影校正

X-ray computed tomographyimage processingimage segmentationthree-dimensional structural quantificationartifact correction

《实验技术与管理》 2026 (10)

1-7,7

国家自然科学基金(42201018)

10.16791/j.cnki.sjg.2026.10.001

评论