Auto-detection of abnormal cells
Biology combined with Info. Tech! Auto-detection of abnormal cells and quick determination of possible DNA damage realized by the KAIT’s newly developed AI-Driven Cell Analysis System; highly expected for genotoxicity testing for finding chromosome disorders
or inspection for pollutants of environmental hazards.
The team from Biomedical Research Center at KAIT developed the auto-detection / classification system of cell disorders in the display images through AI technology, significantly speeding up the genotoxicity testing.
1. Abstracts of the Project
■Designing the software to minimize the time for DNA damage detection in a standardized
genotoxicity testing
■ Artificial Intelligence utilization followed by AI education
■Eliminating the instability in testing derived from the human skill differences and contributing to the enormous workforce reduction.
■Enabling easy access to exam results by simplifying the testing process, and enhancing the use for everyone.
■Determining and inspecting the chemical cause of bio/geo environmental changes and contributing to the study of DNA damages.
The study outcome was published on “Genes and Environment (G&E)”, an academic journal,
The Japanese Environmental Mutagen and Genome Society, and was selected as Featured Article.
2． Background
Micronucleus test is a standardized genotoxicity checking procedure to find DNA damages derived from the inducers of aneuploidy or chromosomal structural aberration. The test procedure are simple and the analysis of the result is easy. However, the procedure still largely depends on human counting; to count the numbers of the whole cells and those embodying the micronuclei through a microscope by the tester with his/her naked eyes. This often affects the outcome due to the differences in the testers’ skills and occasional miscounts. Therefore, the auto-inspecting software of micronucleus has a great impact on this field. Though there is some existing software, many pieces are not feasible because of their high prices or require complex settings since many of them are not intended for this research.
3. The project and outcome
Micronucleus test outcome is calculated based on the number of MN Cells, forming the micronucleus. The procedure requires counting the number of MN Cells in the interphase of 1000 cultured cells through microscopes which takes very long hours, and even longer if the sample increases. To avoid miscounts from human eye counting and shorten the inspection time, different improvement plans have been proposed so far.
Now, the team has developed an application that uses Artificial Intelligence to count micronuclei and inspect cells. The AI has been trained using images of both normal cells and cells with some disorders, allowing it to classify normal and abnormal cell images. The system uses CNN (Convolutional Neural Network), a part of deep learning, to create specialized software for checking micronuclei and cells, making it easy to use.
The application can separate the acridine orange-colored micronuclei images per their RGB channels. The cell nuclei and micronuclei can be found by scaling the G image, while the cytoplasm is done so by recognizing the combined images of R and G. Finally, those cells overlapped with cytoplasm and micronucleus are recognized as micronucleus cells, and then the application displays the number of micronucleus cells as well as the whole cells numbers.
Micronucleus Detecting Application Interface
A) display showing the original image (left) and the image after the analysis
B) name of the image folder
C) the button for engaging the image analysis to display the result
D) the switcher for showing detected images of cytoplasm and micronuclei
E) the switcher for analyzed images
F) the windows indicating the number of analyzed cells and micronuclei
G) adjusters for parameters detecting micronuclei
The system enabled an outcome as good as that of manual counting and was concluded that
it attained the same quality as the manual count, showing the exact number of the whole cells and the micronucleus cells observed through the microscope. Furthermore, the time required for this analysis was shorter than 10% of that for manual operation. Since the quality of counting micronuclei may vary due to cell coloring conditions, the software is equipped with manual parameters for threshold, noise reduction, or binary options, so that the most optimized data could be obtained.
4． Future of the software
The team hopes this application will be used by young scholars or researchers, especially those involved in developing measures for cancer control or environmental pollution.
5．Authors
Takeji Takamura , Faculty of Applied Bioscience
Ayumi Koike, Faculty of Applied Bioscience
Hiromi Yoda ,Faculty of Applied Bioscience
＜ Publication＞
Yoda H, Abe K, Takeo H, Takamura-Enya T, Koike-Takeshita A. Application of image-recognition techniques to automated micronucleus detection in the in vitro micronucleus assay. Genes Environ. 46(1), 11 (2024).
https://doi.org/10.1186/s41021-024-00305-9 selected as Featured Article
＜Research Grants＞
This is a summary from the excerpts of the project of the research of Private Universities Strategic Research Infrastructure Formation Support Project, 2015 to 2019, the Ministry of Education, Culture, Sports, Science and Technology.
＜Glossary＞
*1 Number of chromosomes are larger or fewer than the normal number of found in each species.
*2 The period in the cell cycles other than cell division when synthesis of DNA, RNA, and protein are conducted.
＜Links＞
Published on Genes and Environment (G& E), academic journal of The Japanese Environmental Mutagen and Genome Society
https://doi.org/10.1186/s41021-024-00305-9
▼Inquiries :
Bio-Medical Research Center, Kanagawa Institute of Technology
Ayumi Koike, Professor, Faculty of Applied Bioscience
Keiju Takamura, Professor, Faculty of Applied Bioscience