Editorial: medical imaging modeling,Visual Computing for Industry, Biomedicine, and Art

当前位置： X-MOL 学术 › Vis. Comput. Ind. Biomed. Art › 论文详情

Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)

Editorial: medical imaging modeling
Visual Computing for Industry, Biomedicine, and Art Pub Date : 2019-12-29 , DOI: 10.1186/s42492-019-0037-2
Zhengrong Jerome Liang ₁

Affiliation

Medical imaging can be categorized broadly into applied science or more narrowly into imaging science. As applied science, it encompasses the middle ground between basic science and engineering technology, where a definite practical end is in mind, and where the approach may lead to new discoveries in basic science as well. As imaging science, it explores the nature of visualization.

More specifically, medical imaging visualizes the inner world of the human body noninvasively. Its primary objective is to delineate the structures and map the functions of organs and tissues, based on the principles of physics, mathematics, engineering, computer science, physiology and biology.

Due to the complexities of the human body and the associated signals generation and detection, the tasks of delineating the structures and mapping the functions of the organs and tissues can be very challenging. Modeling or simplifying the complexities to visualize the major attributes of the structures and functions is a commonly adapted strategy. While all models reported in the literature are wrong in terms of describing or representing the complexities, they are very useful and have been contributing to the accumulation of current knowledges about our living body in various conditions.

This special issue calls for papers which would provide some innovative (revolutionary) ideas or models for simplifying the complexities with some preliminary demonstrations and would furthermore provide some insightful discussions about the impacts on and possible future directions of the corresponding fields. The ten papers presented in this special issue can be summarized accordingly.

Computed tomography (CT) is now a widely used imaging modality for screening and diagnosis, emergency medicine, image-guided interventions and monitoring of therapeutic responses because of its excellent capabilities of delineating the structures of the organs and tissues and mapping the contrast material dynamic distributions through the tissues [1]. However, when it is applied to cardiac and pulmonary imaging studies, the motions of heart and lungs complicate the image reconstruction task. The four-dimensional cone-beam CT (4D-CBCT) paper of Zhang, Huang and Wang provides a model of including motion estimation and machine learning to mitigate the motion complexity challenge. Other approaches to address this complexity can be found in the cited references.

For tomographic image reconstruction of a continuous region from limited number of projections around the region, sparse-view problem always exists unless the number of the projections goes to infinity [2]. The problem can be significant in some cases, depending on the applications. The work of Larry Zeng on sparse-view tomography presents a model which addresses the non-linear problem by integrating a non-linear displacement function into a simple linear interpolation. Other models for the sparse view tomography can be found in the cited references, and a more sophisticated model was recently presented for low-dose CT imaging [3].

The beauty of tomographic image reconstruction from projections is seen from the art of filtered back-projection (FBP) reconstruction, which is so far a mathematically-exact solution of the reconstruction in the absence of data noise. However, when the data counts are limited, for example for low-dose CT [4], modeling the count statistics has shown great potential to improve the reconstruction quality. A classic model for Poisson noise statistics is presented by the well-known Expectation-Maximization (EM) algorithm [5, 6]. By theory, the reconstruction problem in limited counts shall be formulated as an optimization operation, given the measured counts, leading to the Bayesian image reconstruction framework [7]. According the Bayes’ law, Bayesian medical image reconstruction remains an unsolved statistical optimization problem because the a priori probability distribution remains a challenging task and, even if an ad hoc prior model is proposed, the associate parameters are remaining as a variable. The paper of Zeng and Li on extension of EM algorithm to Bayesian algorithm provides some insights how these two algorithms are connected.

Positron emission tomography (PET) has many attractive properties as a functional imaging modality to map the tissue heterogeneity at the molecular level, thus called molecular imaging modality [8]. While the radiotracer is labeled at the molecular level of microns (μm) scale, the detection of the radiotracer decay is at the millimeter (mm) scale. Great efforts have been devoted to improve the spatial resolution of localizing the decay sites. One example is the consideration of the time-of-flight (TOF) information of the two 512 keV gamma rays of positron annihilation. The TOF information is associated with the line-of-response (LOR) measurement, and is routinely implemented in an iterative image reconstruction algorithm. The study from Zeng, Li and Huang explores an alternative implementation, where the LOR is back-projected first, followed by filtering in the back-projection domain with an analytical model.

The above four papers focus on image reconstruction, where the ultimate goal is to bring as much information as possible into the reconstructed images to achieve a desired clinical task. For medical diagnosis, imaging the tissue heterogeneity is probably the most important task, because the heterogeneity is a footprint of lesion evolution and ecology, and an indicator of lesion progress and response to intervention [9,10,11]. The tissue heterogeneity has been shown to be effectively characterized by tissue textures [12,13,14]. The tissue textures are entirely dependent on the image contrast. For CT imaging, the image contrast depends entirely on the X-ray energy and, therefore, spectral CT can alter the tissue textures for the task of tissue characterization. The paper of Gao, Shi, Cao, et al., presents a Bayesian image reconstruction of spectral CT framework with enhanced tissue textures, followed by a texture analysis along the spectral dimension for computer aided diagnosis (CADx) of colorectal polyps for colon cancer prevention. This exploratory study is probably the earliest example of focusing the tissue textures into an integrated pipeline from image reconstruction and processing to CADx. In addition, the known normal tissues of muscle, fat, bone, and lungs are incorporated into the Bayesian framework as a priori knowledge and, therefore, the tissue-specific texture prior model is no longer ad hoc prior model.

In the following five papers of this special issue, the focus is shifted from image reconstruction to image processing toward computer-aided detection (CADe) and CADx of abnormalities. As noted in refs. [10, 15], a lesion is embedded in its particular environment. Traditionally, CADx limited the attention on the lesion volume, ignoring its surrounding environment. The model presented by Zheng, Qiu, Aghael, et al. includes both the lesion volume and its surrounding environment. Furthermore, the investigators propose a machine learning strategy to extract the corresponding global features for improved prediction of cancer risk and prognosis. Other models to address the complexity of lesion itself and its surrounding environment can be found in the cited references.

Since the report of radiomics as an integrated lesion descriptor, including image features, experts’ text descriptions and lesion genetics [16], many radiomics features have been proposed. When these radiomics features are input to a machine learning classifier for lesion diagnosis, one or a few features with contributions of more or less variation (due to redundancy and other causes) would accumulate toward a significant impact to the final classification outcome. Evaluating the robustness of each radiomics feature is very necessary. The paper of Cattell, Chen, and Huang present an example of research efforts to address this important topic in the radiomics field.

As mentioned above in the paper of Gao, Shi, Cao, et al., tissue texture patterns effectively represent the tissue heterogeneity and, therefore, are considered as imaging biomarkers. Many texture descriptors (also called texture features in terms of quantitative measures or mathematical expression) are also investigated within the radiomics framework. Robustness of texture descriptors is an important research topic as shown by the work of Cattell, Chen, and Huang. The paper of Cao, Pomeroy, Gao, et al. takes the gray-level co-occurrence (GLCM) texture descriptor as an example to explore the possible causes of the variations, and provides a solution to minimize the variations by adaptive machine learning. More importantly, this paper of Cao, Pomeroy, Gao, et al. takes the GLCM as a texture image and explores the multiscale texture sampling opportunity in a similar manner as multiscale image analysis.

As mentioned above, imaging the tissue heterogeneity is probably the most important task in medical diagnosis, image-guided intervention, and treatment response quantification [9,10,11,12,13,14]. Thus, multimodal imaging becomes a desirable approach to acquire multiscale, multispectral, and multiple-source images. The paper of Ma, Lu, Wang, et al. presents another very valuable imaging modality, called hyperspectral imaging, to address the challenge of differentiating tumor and benign tissues for optimal surgical purpose.

Lung cancer remains the leading cause of cancer-related deaths. Detecting the lung nodules, the precursor of lung cancer, is the primary consideration to prevent and treat the disease early. Currently the task of detecting the nodules is performed by low-dose CT (LdCT) scanning, followed by radiologists’ interpretation of the reconstructed LdCT chest images. Artificial intelligence (AI) has been traditionally applied to the reconstructed images to mimic the experts’ interpretation. During the image reconstruction processing from acquired raw data to chest images, additional variables may be introduced, which can compromise the detection task. Therefore, a logic strategy to accomplish the detection task is taking the AI-enabled learning during the entire process from the raw data to the reconstructed images. The paper of Gao, Tan, Liang, et al. presents such a logic strategy, which has shown significant improvement over the traditional reconstructed image-focused approaches.

In summary, the ten papers in this special issue cover a wide range of modeling strategies to address the complexity in medical diagnosis, image-guided intervention, and treatment response quantification, which is usually impossible to solve without innovative modeling strategy.

1.
Buzug TM (2008) Computed tomography: from photon statistics to modern cone-beam CT. Springer-Verlag, Berlin Germany
Google Scholar
2.
Tan J, Gao Y, Huo Y, Li L, Liang Z (2019) Sharpness preserved sinogram synthesis using convolutional neural network for sparse-view CT imaging. Paper presented at the SPIE Medical Imaging, Town and Country Resort & Convention Center, San Diego, 16-21 February 2019. https://doi.org/10.1117/12.2512894
Google Scholar
3.
Liu Y, Liang Z, Ma J, Lu H, Wang K, Zhang H et al (2014) Total variation-stokes strategy for sparse-view X-ray CT image reconstruction. IEEE Trans Med Imaging 33(3):749–763. https://doi.org/10.1109/TMI.2013.2295738
Article Google Scholar
4.
Liang Z, La Riviere P, El Fakhri G, Glick SJ, Siewerdsen JH (2017) Low-Dose CT: What has been done, and what challenges remain. Guest Editorial in IEEE Transactions on Medical Imaging 36(12):2409–2416. https://doi.org/10.1109/TMI.2017.2768978
Article Google Scholar
5.
Shepp LA, Vardi Y (1982) Maximum likelihood reconstruction for emission tomography. IEEE Trans Med Imaging 1(1):113–122. https://doi.org/10.1109/TMI.1982.4307558
Article Google Scholar
6.
Lange K, Carson R (1984) EM reconstruction algorithms for emission and transmission tomography. J Comput Assist Tomogr 8(2):306–316
Google Scholar
7.
Liang Z (1987) Bayesian image processing of data from constrained source distributions, PhD Dissertation. The City University of New York, New York
Google Scholar
8.
Tanaka E (1987) Recent Progress on single photon and positron emission tomography - from detectors to algorithms. IEEE Trans Nucl Sci 34(1):313–320. https://doi.org/10.1109/TNS.1987.4337355
Article Google Scholar
9.
Chicklore S, Goh V, Siddique M, Roy A, Marsden PK, Cook GJR (2013) Quantifying tumor heterogeneity in 18F-FDG PET/CT imaging by texture analysis. Eur J Nucl Med Mol Imaging 40(1):133–140. https://doi.org/10.1007/s00259-012-2247-0
Article Google Scholar
10.
Gatenby RA, Grove O, Gillies RJ (2013) Quantitative imaging in cancer evolution and ecology. Radiology 269(1):8–15. https://doi.org/10.1148/radiol.13122697
Article Google Scholar
11.
Bashir U, Siddique M, Mclean E, Goh V, Cook GJ (2016) Imaging heterogeneity in lung cancer: techniques, applications, and challenges. Am J Roentgenol 207(3):534–543. https://doi.org/10.2214/AJR.15.15864
Article Google Scholar
12.
Buvat I, Orlhac F, Soussan M (2015) Tumor texture analysis in PET: where do we stand? J Nucl Med 56(11):1642–1644. https://doi.org/10.2967/jnumed.115.163469
Article Google Scholar
13.
O'Connor JPB, Jackson A, Asselin MC, Buckley DL, Parker GJM, Jayson GC (2008) Quantitative imaging biomarkers in the clinical development of targeted therapeutics: current and future perspectives. Lancet Oncol 9(8):766–776. https://doi.org/10.1016/S1470-2045(08)70196-7
Article Google Scholar
14.
O'Connor JPB, Rose CJ, Waterton JC, Carano RAD, Parker GJM, Jackson A (2015) Imaging intra tumor heterogeneity: role in therapy response, resistance, and clinical outcome. Clin Cancer Res 21(2):249–257. https://doi.org/10.1158/1078-0432.CCR-14-0990
Article Google Scholar
15.
Pomeroy M, Abbasi A, Baker K, Barish M, Pickhardt P, Zhang G et al (2017) Texture feature analysis of neighboring colon wall for colorectal polyp classification. In: Paper presented at the 2017 IEEE nuclear science symposium and medical imaging conference. IEEE, Atlanta. https://doi.org/10.1109/NSSMIC.2017.8532773
Google Scholar
16.
Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RGPM, Granton P et al (2012) Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 48(4):441–446. https://doi.org/10.1016/j.ejca.2011.11.036
Article Google Scholar

Download references

Affiliations

Laboratory for Imaging Research and Informatics (IRIS), State University of New York, Stony Brook, New York, 11794, USA
- Zhengrong Jerome Liang

Authors

Zhengrong Jerome LiangView author publicationsYou can also search for this author in
- PubMed
- Google Scholar

Contributions

ZJL read and approved the final manuscript.

Corresponding author

Correspondence to Zhengrong Jerome Liang.

Competing interests

The author declares that he has no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Reprints and Permissions

Cite this article

Liang, Z.J. Editorial: medical imaging modeling. Vis. Comput. Ind. Biomed. Art 2, 26 (2019). https://doi.org/10.1186/s42492-019-0037-2

Download citation

Received: 05 November 2019
Accepted: 06 November 2019
Published: 29 December 2019
DOI: https://doi.org/10.1186/s42492-019-0037-2

中文翻译：

社论：医学影像建模

医学成像可以广义地分类为应用科学，也可以狭义地分类为成像科学。作为应用科学，它涵盖了基础科学和工程技术之间的中间地带，其中牢记着一定的实践目的，并且该方法也可能导致基础科学方面的新发现。作为成像科学，它探索了可视化的本质。

更具体地，医学成像无创地可视化人体的内部世界。它的主要目标是根据物理学，数学，工程学，计算机科学，生理学和生物学原理，描绘器官和组织的结构并绘制功能图。

由于人体的复杂性以及相关信号的产生和检测，描绘器官的结构和绘制器官和组织功能的任务非常艰巨。建模或简化复杂性以可视化结构和功能的主要属性是一种通常适用的策略。尽管文献中报道的所有模型在描述或表示复杂性方面都是错误的，但是它们非常有用，并且有助于在各种情况下积累有关我们生物体的当前知识。

本期特刊要求发表一些论文，这些论文将提供一些创新的（革命性的）思想或模型，以通过一些初步的演示来简化复杂性，并且还将提供一些对有关领域的影响和未来方向的深刻讨论。本期特刊提出的十篇论文可以作相应总结。

计算机断层扫描（CT）由于其描绘器官和组织结构以及绘制对比材料动态分布的出色能力，现在已成为广泛用于筛查和诊断，急诊医学，图像指导的干预措施和治疗反应监测的成像方式。通过组织[1]。但是，当将其应用于心脏和肺部成像研究时，心脏和肺部的运动会使图像重建任务变得复杂。Zhang，Huang和Wang的4维锥束CT（4D-CBCT）论文提供了一个包含运动估计和机器学习的模型，以减轻运动复杂性的挑战。可以在引用的参考文献中找到解决这种复杂性的其他方法。

对于从区域周围有限数量的投影重建连续区域的断层图像，除非投影的数量达到无穷大，否则始终存在稀疏视图问题[2]。在某些情况下，取决于应用程序，该问题可能很严重。Larry Zeng在稀疏断层层析成像方面的工作提出了一个模型，该模型通过将非线性位移函数集成到简单的线性插值中来解决非线性问题。可以在引用的参考文献中找到其他用于稀疏视图层析成像的模型，并且最近提出了一种用于低剂量CT成像的更复杂的模型[3]。

从滤波的反投影（FBP）重建技术中可以看到从投影重建断层图像的美妙之处，到目前为止，这是在没有数据噪声的情况下重建的数学上精确的解决方案。然而，当数据计数受到限制时，例如对于低剂量CT [4]，对计数统计数据进行建模已显示出巨大的潜力，可以改善重建质量。泊松噪声统计的经典模型由著名的期望最大化（EM）算法提出[5，6]。从理论上讲，在给定测量值的情况下，应将有限数量的重建问题公式化为优化操作，从而导致贝叶斯图像重建框架[7]。根据贝叶斯定律，贝叶斯医学图像重建仍然是一个尚未解决的统计优化问题，因为先验概率分布仍然是一项艰巨的任务，即使提出了临时的先验模型，相关参数仍将保留为变量。Zeng和Li在将EM算法扩展到贝叶斯算法方面的论文提供了一些有关这两种算法如何连接的见解。

正电子发射断层扫描（PET）作为功能成像方式具有许多吸引人的特性，可以在分子水平上绘制组织异质性，因此称为分子成像方式[8]。虽然放射性示踪剂的标记分子水平为微米（μm），但放射性示踪剂衰减的检测值为毫米（mm）尺度。已经作出巨大的努力来提高定位衰减点的空间分辨率。一个示例是考虑正电子an灭的两条512 keV伽马射线的飞行时间（TOF）信息。TOF信息与响应线（LOR）测量相关联，并且通常在迭代图像重建算法中实现。Zeng，Li和Huang进行的研究探索了另一种实施方式，即先对LOR进行背投，

以上四篇论文集中在图像重建上，其最终目标是将尽可能多的信息带入重建的图像中，以实现所需的临床任务。对于医学诊断，对组织异质性进行成像可能是最重要的任务，因为异质性是病变发展和生态的足迹，是病变进展和对干预反应的指标[9,10,11]。组织异质性已被证明可以有效地表征组织质地[12,13,14]。组织纹理完全取决于图像对比度。对于CT成像，图像对比度完全取决于X射线能量，因此，光谱CT可以更改组织纹理以进行组织表征。高石，曹等人的论文，提出了具有增强的组织纹理的光谱CT框架的贝叶斯图像重建，然后沿光谱维度进行纹理分析，以用于结肠直肠息肉的计算机辅助诊断（CADx）预防结肠癌。这项探索性研究可能是将组织纹理集中到从图像重建和处理到CADx的集成管道中的最早示例。另外，将肌肉，脂肪，骨骼和肺部的已知正常组织作为先验知识合并到贝叶斯框架中，因此，组织特定纹理的先验模型不再是即席先验模型。这项探索性研究可能是将组织纹理集中到从图像重建和处理到CADx的集成管道中的最早示例。另外，将肌肉，脂肪，骨骼和肺部的已知正常组织作为先验知识合并到贝叶斯框架中，因此，组织特定纹理的先验模型不再是即席先验模型。这项探索性研究可能是将组织纹理集中到从图像重建和处理到CADx的集成管道中的最早示例。另外，将肌肉，脂肪，骨骼和肺部的已知正常组织作为先验知识合并到贝叶斯框架中，因此，组织特定纹理的先验模型不再是即席先验模型。

在此特刊的以下五篇论文中，重点从图像重建转向图像处理，转向计算机辅助检测（CADe）和异常CADx。如参考文献所述。[10，15]，病变嵌入其特定的环境。传统上，CADx忽略了周围的环境，将注意力集中在病变体积上。Zheng，Qiu，Aghael等人提出的模型。包括病变体积及其周围环境。此外，研究人员提出了一种机器学习策略，以提取相应的全局特征，以改善对癌症风险和预后的预测。可以在引用的参考文献中找到解决病变本身及其周围环境复杂性的其他模型。

自从放射线学作为一个综合的病变描述报告以来，包括图像特征，专家的文字描述和病变遗传学[16]，已经提出了许多放射线学特征。当将这些放射线学特征输入到机器学习分类器中以进行病变诊断时，一个或几个具有或多或少的变化贡献的特征（由于冗余和其他原因）将对最终分类结果产生重大影响。评估每个放射线学功能的鲁棒性是非常必要的。卡特尔（Cattell），陈（Chen）和黄（Guang）的论文提供了一个研究成果的例子，以解决放射学领域的这一重要课题。

如上面在Gao，Shi，Cao等人的论文中所述，组织纹理图案有效地代表了组织异质性，因此被认为是成像生物标记。在radimics框架内还研究了许多纹理描述符（在定量度量或数学表达式方面也称为纹理特征）。正如Cattell，Chen和Huang的工作所示，纹理描述符的鲁棒性是一个重要的研究课题。曹，波莫罗，高等人的论文。以灰度共现（GLCM）纹理描述符为例，探讨了引起变化的可能原因，并提供了一种通过自适应机器学习将变化最小化的解决方案。更重要的是，Cao，Pomeroy，Gao等人的论文。

如上所述，对组织异质性成像可能是医学诊断，图像引导干预和治疗反应量化中最重要的任务[9,10,11,12,13,14]。因此，多峰成像成为获取多尺度，多光谱和多源图像的理想方法。Ma，Lu，Wang等人的论文。提出了另一种非常有价值的成像方式，称为高光谱成像，以解决区分肿瘤和良性组织以达到最佳手术目的的挑战。

肺癌仍然是与癌症有关的死亡的主要原因。发现肺结节（肺癌的前兆）是及早预防和治疗该疾病的主要考虑因素。当前，通过低剂量CT（LdCT）扫描执行检测结节的任务，然后放射科医生对重建的LdCT胸部图像进行解释。传统上，将人工智能（AI）应用于重建图像以模仿专家的解释。在从获取的原始数据到胸部图像的图像重建过程中，可能会引入其他变量，这可能会损害检测任务。因此，完成检测任务的逻辑策略是在从原始数据到重建图像的整个过程中进行AI启用学习。高，谭，梁等人的论文。

总而言之，本期特刊的十篇论文涵盖了广泛的建模策略，以解决医学诊断，图像指导的干预和治疗反应量化中的复杂性，而如果没有创新的建模策略，通常是无法解决的。

1。
Buzug TM（2008）计算机断层扫描：从光子统计到现代锥束CT。施普林格出版社，柏林德国
谷歌学术
2。
谭J，高Y，霍Y，李L，梁Z（2019）使用卷积神经网络进行稀疏度CT成像的清晰度保留正弦图合成。该论文于2019年2月16日至21日在圣地亚哥的SPIE医学影像，城镇和乡村度假及会议中心发表.https：//doi.org/10.1117/12.2512894
谷歌学术
3。
Liu Y，Liang Z，Ma J，Lu H，Wang K，Zhang H等（2014）稀疏视图X射线CT图像重建的总变幅斯托克斯策略。IEEE Trans Med影像33（3）：749–763。https://doi.org/10.1109/TMI.2013.2295738
文章Google学术搜索
4。
Liang Z，La Riviere P，El Fakhri G，Glick SJ，Siewerdsen JH（2017）低剂量CT：已完成的工作以及仍然面临的挑战。在IEEE Transactions on Medical Imaging上的客座社论36（12）：2409-2416。https://doi.org/10.1109/TMI.2017.2768978
文章Google学术搜索
5，
Shepp LA，Vardi Y（1982）放射线断层摄影的最大似然重建。IEEE Trans Med成像1（1）：113–122。https://doi.org/10.1109/TMI.1982.4307558
文章Google学术搜索
6。
Lange K，Carson R（1984）EM重建算法用于发射和透射层析成像。J Comput Assist Tomogr 8（2）：306–316
谷歌学术
7。
Liang Z（1987）约束源分布数据的贝叶斯图像处理，博士学位论文。纽约城市大学，纽约
谷歌学术
8。
田中E（1987）单光子和正电子发射断层扫描的最新进展-从探测器到算法。IEEE Trans Nucl Sci 34（1）：313–320。https://doi.org/10.1109/TNS.1987.4337355
文章Google学术搜索
9。
Chicklore S，Goh V，Siddique M，Roy A，Marsden PK，Cook GJR（2013）通过质构分析定量18F-FDG PET / CT成像中的肿瘤异质性。Eur J Nucl医学分子影像学40（1）：133–140。https://doi.org/10.1007/s00259-012-2247-0
文章Google学术搜索
10。
Gatenby RA，Grove O，Gillies RJ（2013）癌症进化和生态学中的定量成像。放射学269（1）：8-15。https://doi.org/10.1148/radiol.13122697
文章Google学术搜索
11。
Bashir U，Siddique M，Mclean E，Goh V，Cook GJ（2016）肺癌成像异质性：技术，应用和挑战。Am J Roentgenol 207（3）：534–543。https://doi.org/10.2214/AJR.15.15864
文章Google学术搜索
12
Buvat I，Orlhac F，Soussan M（2015）PET中的肿瘤纹理分析：我们站在哪里？J Nucl Med 56（11）：1642-1644。https://doi.org/10.2967/jnumed.115.163469
文章Google学术搜索
13
O'Connor JPB，Jackson A，Asselin MC，Buckley DL，Parker GJM，Jayson GC（2008）靶向治疗药物临床开发中的定量成像生物标志物：当前和未来观点。Lancet Oncol 9（8）：766-776。https://doi.org/10.1016/S1470-2045(08)70196-7
文章Google学术搜索
14。
O'Connor JPB，Rose CJ，Waterton JC，Carano RAD，Parker GJM，Jackson A（2015）成像内肿瘤异质性：在治疗反应，耐药性和临床结果中的作用。临床癌症研究杂志21（2）：249-257。https://doi.org/10.1158/1078-0432.CCR-14-0990
文章Google学术搜索
15
Pomeroy M，Abbasi A，Baker K，Barish M，Pickhardt P，Zhang G等人（2017）结直肠息肉分类的邻近结肠壁纹理特征分析。在：在2017 IEEE核科学研讨会和医学影像会议上发表的论文。IEEE，亚特兰大。https://doi.org/10.1109/NSSMIC.2017.8532773
谷歌学术
16。
Lambin P，Rios-Velazquez E，Leijenaar R，Carvalho S，van Stiphout RGPM，Granton P等人（2012）Radiomics：使用高级特征分析从医学图像中提取更多信息。Eur J Cancer 48（4）：441–446。https://doi.org/10.1016/j.ejca.2011.11.036
文章Google学术搜索