当前位置: X-MOL 学术Cytom. Part A › 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.)
Make them open and more about image cytometry
Cytometry Part A ( IF 2.5 ) Pub Date : 2021-06-22 , DOI: 10.1002/cyto.a.24476
Attila Tárnok 1, 2, 3
Affiliation  

As I write the June editorial, CYTO Interactive is taking place virtually. Compared with the CYTO 2020, virtual by necessity, I definitely found this year's format more attractive. Discussion forums, break out session and various “Lounges” created a feeling of being there and discussing in person. Even then, there is a general saturation of these exclusively remote meetings, and it is no surprise that people long to get back to normal at least partially. A real discussion with a real person is different.

In that same vein, reflecting the real cytometry community, Cytometry A is looking to insert transparency in the editorial process, particularly as it pertains to the anonymity of peer-reviewers and peer-reviews. To that effect, Cytometry Part A will implement an Open Peer Review process in the near future. Many journals have already adopted some form of open peer review. For Cytometry Part A, author and reviewer participation is voluntary, and they will be able to opt in or out. If they agree, then with the finally accepted paper will be amended by the reviewers' comments and (if selected) also by then name of the reviewer. This is an active step forward to facilitate open science and transparency to the readers and the authors.

In this July issue I have compiled six manuscripts around the topic of image analysis for cytometric cell analysis, an area that has been already in the focus of several special issues of Cytometry Part A.

Cytotoxicity assays are in the focus of the first two manuscripts in this line. Cytotoxicity tests are common all over the field of biomedicine, including drug development, cancer therapy, and beyond. Wu et al. (pp. 678–686) combined traditional flow cytometric cytotoxicity assay using CFSE and HOE 33258 with imaging cytometry in flow. The authors could improve the detection specificity of the assay and reduce incubation time to 15 s, enabling it for higher-throughput applications. Wang et al. (pp. 687–695) used image cytometry to improve, CAR-T cell production and quality. CAR-T cells are genetically engineered autologous T-cells with a chimeric antigen receptor. CAR cells are the tool and hope for otherwise incurable malignancies. By image cytometry transduction efficiency, cell proliferation and general and specific cell killing efficiency were tested. Image cytometry yielded comparable results to with standard flow cytometry and/or 51Cr release-assay but required substantially less cells and was therefore more efficient.

Nuclear feature analysis from microscopic images involving artificial intelligence could support manual analysis of complex specimens and reduce error rate of manual analysis [1]. Lee et al. (pp. 696–704) were focusing their work on nuclear texture analysis in breast cancer cells in human tumors. Nuclear texture features are of relevance in diagnosis and grading of tumor cells. After cancer cell nuclei segmentation from images of H&E labeled tissues texture features were extracted by image analysis. Images were then classified using machine learning with support vector machine and k-nearest neighbor algorithms. For the trained algorithms the correct classification was highest when different nuclear features were combined and yielded over 90% accuracy.

Woloshuk et al. (pp. 705–719) involved Deep Learning based on Convolutional Neural Network to classify three-dimensional (3D) segmented images of cell nuclei of the human kidney based on nuclear shape features. After confocal tissue imaging nuclei were segmented by tissue cytometry [2] and manually classified into eight different cell types by experts. After training the algorithm, accuracy of nucleus type prediction that was based exclusively on nuclear features was higher for 3D rendered nuclei than from two-dimensional (2D) images. Accuracy increased to over 80% if additional information on nuclear environment such as surrounding tissue was provided.

Urine cell cytology for bladder cancer diagnosis is addressed by the two last papers in this selection. Both used computer-based visioning, recognition, and classification of normal and malignant cells in the human urine. Lv et al. (pp. 720–729) developed a microfluidic chip system where urine cells are microencapsulated in oil droplets and imaged. This is followed by automated image segmentation. The results demonstrate that oil encapsulation is superior to other preparation techniques for reliable cell recognition and classification. Awan et al. (pp. 730–740) used datasets from cytological images of patients. The cells were classified by expert observers and then used as training or test data using two deep learning approaches. Following cell identification and segmentation the power of discrimination by the trained classifiers displayed ROC values above 0.8.

In conclusion, combination flow cytometry with image cytometry and bioinformatics is promising for automating and improving diagnosis. Still further clinical studies are needed to prove their value in the real world.



中文翻译:

让他们敞开心扉,了解更多关于图像细胞术的信息

在我撰写六月社论时,CYTO Interactive 正在虚拟进行。与必须虚拟的 CYTO 2020 相比,我绝对发现今年的格式更具吸引力。讨论论坛、分组会议和各种“休息室”营造了一种身临其境并亲自讨论的感觉。即便如此,这些完全远程会议也普遍饱和,人们渴望至少部分恢复正常也就不足为奇了。与真实的人进行真正的讨论是不同的。

同样,反映真正的细胞学社区,Cytometry A 希望在编辑过程中增加透明度,特别是因为它与同行评审员和同行评审的匿名性有关。为此,Cytometry Part A 将在不久的将来实施开放同行评审流程。许多期刊已经采用了某种形式的开放同行评审。对于 Cytometry Part A,作者和审稿人的参与是自愿的,他们可以选择加入或退出。如果他们同意,那么最终接受的论文将根据审稿人的意见进行修改,并且(如果被选中)还包括当时的审稿人姓名。这是向读者和作者促进开放科学和透明度迈出的积极一步。

在今年 7 月的期刊中,我围绕细胞计数细胞分析的图像分析这一主题编写了六篇手稿,该领域已经成为细胞计数 A 部分几期特刊的焦点。

细胞毒性试验是这一行前两份手稿的重点。细胞毒性测试在整个生物医学领域都很常见,包括药物开发、癌症治疗等。吴等人。(pp. 678–686) 将使用 CFSE 和 HOE 33258 的传统流式细胞术细胞毒性测定与流式细胞术成像相结合。作者可以提高检测的检测特异性,并将孵育时间缩短至 15 秒,使其能够用于更高通量的应用。王等人。(pp. 687–695) 使用图像细胞计数来改善 CAR-T 细胞的生产和质量。CAR-T 细胞是具有嵌合抗原受体的基因工程自体 T 细胞。CAR 细胞是治疗其他无法治愈的恶性肿瘤的工具和希望。通过图像细胞仪转导效率、细胞增殖和一般和特异性细胞杀伤效率进行测试。51 Cr 释放测定,但需要的细胞显着减少,因此效率更高。

涉及人工智能的显微图像核特征分析可以支持复杂标本的人工分析,降低人工分析的错误率[ 1 ]。李等人。(pp. 696–704) 将他们的工作重点放在人类肿瘤中乳腺癌细胞的核结构分析上。核纹理特征与肿瘤细胞的诊断和分级相关。在从 H&E 标记的组织图像中分割出癌细胞核后,通过图像分析提取了纹理特征。然后使用支持向量机和 k-最近邻算法的机器学习对图像进行分类。对于经过训练的算法,当不同的核特征组合时正确分类最高,并且产生超过 90% 的准确率。

沃洛舒克等人。(pp. 705–719) 涉及基于卷积神经网络的深度学习,以根据核形状特征对人体肾脏细胞核的三维 (3D) 分割图像进行分类。共聚焦组织成像后,细胞核通过组织细胞术 [ 2 ]进行分割,并由专家手动分类为八种不同的细胞类型。在对算法进行训练后,仅基于核特征的核类型预测对于 3D 渲染核的准确度高于二维 (2D) 图像。如果提供有关核环境(例如周围组织)的附加信息,则准确度将提高到 80% 以上。

尿细胞学本选择中的最后两篇论文讨论了膀胱癌诊断。两者都使用基于计算机的视觉、识别和人类尿液中正常和恶性细胞的分类。吕等人。(pp. 720–729) 开发了一种微流控芯片系统,其中尿细胞被微封装在油滴中并成像。然后是自动图像分割。结果表明,油包封在可靠的细胞识别和分类方面优于其他制备技术。阿万等人。(pp. 730–740) 使用来自患者细胞学图像的数据集。这些细胞由专家观察员分类,然后使用两种深度学习方法用作训练或测试数据。在细胞识别和分割之后,经过训练的分类器的区分能力显示 ROC 值高于 0.8。

总之,流式细胞术与图像细胞术和生物信息学的结合有望实现自动化和改进诊断。还需要进一步的临床研究来证明它们在现实世界中的价值。

更新日期:2021-07-01
down
wechat
bug