摘要
图像质量的客观评价具有片面性,只能评估图像质量的某方面;主观评价具有不可重复性,受外部以及观察者本身因素影响,无法确保评价结果的可靠性。本文提出一种综合评价图像质量的方法。对图像数据同时进行主观和客观评价,并将主观评分映射到客观指标。由此,用客观指标综合表征图像的主观和客观评价结果,使得能够被准确定义和计算的客观指标具有临床诊断意义。对于客观评价,计算CNR作为图像的客观指标。对于主观评价,由于主观评分标准的分数定义模糊,提出一种基于图像比较的主观评价方法。该方法通过比较两张图像的质量好坏,根据比较结果对图像质量由好到坏降序排序。邀请未参与图像比较的医生按照主观评分标准划分图像序列,得到图像的主观评分。每张图像的CNR和主观评分一一对应,统计得到主观评分对应的CNR范围(CNR均值±CNR标准差)。结果表明,主观评价的一致性基本高于0. 6,说明主观评价的统计处理是有意义的。随着主观评分下降,对应的CNR基本符合下降趋势,说明图像的客观和主观评价呈单调相关性,可以将主观评分映射到客观指标。不同解剖结构在大小和构成上有一定差异,它们主观评分映射的CNR范围不尽相同,从而可以根据解剖结构对图像质量进行优化。本文将主观评分映射到具体的CNR值,使得客观指标不仅能够定量表示图像质量,而且具有临床辅助意义。
Objective image quality evaluation is one-sided,and can only evaluate certain aspects of image quality.Subjective assessment is non-repeatable,and might be affected by external and internal factors,thereby reducing its reliability.In this work,we proposes a comprehensive image quality evaluation method.Subjective and objective image evaluation are performed simultaneously,and subjective scores are mapped to objective indicators.W e combine objective indicators and subjective scores to make the objective indicators that can be accurately defined and calculated have clinical diagnostic significance.Both objective evaluation and subjective assessment are performed on each image data.For objective evaluation,CNR is calculated as the objective index of image quality.For subjective assessment,because the definition of the subjective score criteria is ambiguous,we proposed a human observer method based on image comparison by simply telling which one is better from an image pair.The consistency of subjective assessment is higher than 0.6,indicating that the process is statically reasonable.All images are sorted in a descending order according to the comparison results.Doctors who are not involved in image comparison are invited to decide the boundary of scores(1-5)for the image sequence.The distribution of CNR(CNR mean±CNR standard deviation)corresponding to each subjective score is studied.CNR decreases as the subjective score decreases,indicating that objective evaluation is monotonously correlated to subjective assessment,and the subjective score can be mapped to the objective index.Different anatomical structures are of different size and composition.Their subjective scores correspond to different CNR distribution,so that image quality shall be optimized according to anatomical structures.This work maps subjective scores to specific CNR values so that CNR can not only be used to quantitatively express image quality but also be clinically meaningful.
作者
郭潇月
邢宇翔
张丽
谢晓艳
张康平
GUO Xiaoyue;XING Yuxiang;ZHANG Li;XIE Xiaoyan;ZHANG Kangping(Department of Engineering Physics,Tsinghua University,Beijing 100084,China;Key Laboratory of Particle&Radiation Imaging(Tsinghua University),Ministry of Education,Beijing 100084,China;Peking University Hospital of stomatology,Radiology Department,Beijing 100081,China;LargeV Instrument Corporation Limited,Beijing 100084,China)
出处
《中国体视学与图像分析》
2019年第4期371-379,共9页
Chinese Journal of Stereology and Image Analysis
基金
国家自然科学基金(No.11435007)
国家自然科学基金(No.61527807)
国家重点研发计划(No.2016YFB101203)。
关键词
CBCT
客观评价
主观评价
CBCT
objective image quality evaluation
human observer
