Language：English   Publishing type：Research paper (scientific journal)  

Biological morphologies of cells and tissues represent their physiological and pathological conditions. The importance of quantitative assessment of morphological information has been highly recognized in clinical diagnosis and therapeutic strategies. In this study, we used a supervised machine learning algorithm wndchrm to classify hematoxylin and eosin (H&E)‐stained images of human gastric cancer tissues. This analysis distinguished between noncancer and cancer tissues with different histological grades. We then classified the H&E‐stained images by expression levels of cancer‐associated nuclear ATF7IP/MCAF1 and membranous PD‐L1 proteins using immunohistochemistry of serial sections. Interestingly, classes with low and high expressions of each protein exhibited significant morphological dissimilarity in H&E images. These results indicated that morphological features in cancer tissues are correlated with expression of specific cancer‐associated proteins, suggesting the usefulness of biomolecular‐based morphological classification.

DOI： DOI: 10.1002/cam4.2885

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1080/19491034.2016.1149664

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1038/srep06996

Event date： 2019.12

Language：Japanese  

Venue：福岡国際会議場   Country：Japan  

Accumulation of genetic and epigenetic abnormalities in cancer leads to dysregulation of gene expression, which may affect dynamic changes of cell and tissue formation. It is critical to understand morphological and molecular features of cells/tissues for cancer diagnosis and therapeutic application.
Hematoxylin and eosin (H&E) stain is routinely utilized for pathological assessment of cancer, and it provides essential information of morphological contexts underlying various molecular events. In this study, we quantitatively evaluated morphological changes of H&E images in gastric cancer tissues using machine-learning algorithm, wndchrm. This analysis revealed that both hematoxylin (nuclear) and eosin (cytoplasmic) images contain morphological features enough to distinguish cancer tissues from normal tissues. We then performed a fact-driven analysis, which is the classification of H&E images by expression levels of cancer-associated nuclear ATF7IP/MCAF1 or membranous PD-L1 protein, to test a correlation between morphological changes and molecular expressions. The results showed high classification performance (90% for ATF7IP/MCAF1 and 86% for PD-L1), based on the high or low expression level of each antigen.
These results show that morphological differences in H&E images are precisely detected by the machine-learning method, and our fact-driven analysis propose an objective identification to assess a biological link between morphologies and molecular expression patterns.

Event date： 2019.11

Language：English  

Venue：ANAクラウンプラザホテル岡山   Country：Japan  

Dynamic changes of cell morphology in cancer tissue reflect accumulation of genetic and epigenetic abnormalities. H&E slide is a conventional tool in pathological diagnosis and provide useful information of a morphological context underlying various molecular events. It is important for cancer therapy and diagnosis to understand the molecular mechanism behind morphological changes, however, little is known.
Here, we quantified morphological dissimilarity of H&E images of gastric cancer with machine-learning algorithm, wndchrm. First, we verified that wndchrm algorithm reproduced cancer classification and differentiated grades based on human vision with acceptable accuracy. We next found that both hematoxylin (nuclear) and eosin (cytoplasmic) images equally contain morphological features enough to distinguish cancer grades. To investigate links between morphological change and molecular expression, we performed a fact-driven analysis, an image classification of H&E corresponding to expression level of ATF7IP/MCAF1 (nuclear protein) or PD-L1 (non-nuclear protein), resulting in high classification performance (90% for ATF7IP/MCAF1; 86% for PD-L1).
Our data shows that quantitative evaluation of H&E images by machine-learning algorithm accurately detected morphological differences and propose the usefulness of machine-learning algorithm as a tool to elucidate a link between morphologies and molecular expression patterns.

Event date： 2019.9

Language：Japanese  

Venue：国立京都国際会館   Country：Japan  

Morphologies of cells and tissues correlate with physiological status reflecting cumulative information of genetic and epigenetic abnormalities. Although its elucidation may give advantages for cancer therapy and diagnosis, molecular link between the morphologies and the abnormalities is largely unknown.
Here, we utilized machine-learning algorithm, wndchrm, to analyze the morphological changes of gastric cancer tissues. We succeeded to quantify the morphological information of tissues, and also found that both hematoxylin (nuclear) and eosin (cytoplasmic) images equally contain morphological features enough to distinguish cancer grades. We next performed a fact-driven analysis, an image classification of H&E-stained slides based on expression level of ATF7IP/MCAF1 (nuclear protein) or PD-L1 (non-nuclear protein), resulting in high classification performance (90% for ATF7IP/MCAF1; 88% for PD-L1).
Our data showed that quantification of morphologies objectively interpreted morphological changes of the tissue images and it might be a potential tool for elucidation of the molecular mechanisms behind morphological changes.

Event date： 2018.1

Language：Others  

Venue：Kumamoto City International Center, Japan   Country：Japan  

Structure and function of cells and tissues are closely correlated in physiology and pathology, together with genetic and epigenetic regulation. For example, in differentiating cells and cancer cells, we can observe characteristic morphological features. However, molecular mechanisms behind these morphologies remain unclear because of our qualitative and descriptive assessments. Recently, automated image analysis has been developed and widely used for biomedical researches. A supervised machine learning algorithm, wndchrm (weighted neighbor distances using a compound hierarchy of algorithms representing morphology), is useful for automated image classification and quantification. We previously reported that wndchrm discriminated between completely multipotent iPS cells and incomplete non-iPS cells by measuring the morphological differences of the colony formation. In the present study, we investigated normal and cancer tissue morphologies with the wndchrm. We collected a large number of HE-stained tissue images from gastric cancer patients using tissue microarrays, and then asked whether wndchrm can discriminate morphological differences between normal and cancer tissues with malignant grades. Cancer tissue images were classified with 91% sensitivity and 99% specificity. When the classification was performed with cell nuclear region in the tissues, we obtained 87.5% sensitivity and 67.5% specificity. Our data showed that wndchrm successfully detected morphological differences between cancer and normal tissues, and that nuclear changes may contribute to the classifications, as is the nuclear atypia in routine pathological examinations. Collectively, computational image analyses are likely to contribute to morphological assessments and future clinical applications.

Event date： 2017.9

Language：Japanese  

Venue：パシフィコ横浜   Country：Japan  

Audience：Infants,　Schoolchildren,　Junior students,　High school students

Type：Seminar, workshop

Audience：General,　Scientific,　Company,　Civic organization,　Governmental agency

Type：Lecture

Audience：General,　Scientific,　Company,　Civic organization,　Governmental agency

Type：Other