Scientists offer dataset of cancer cell images for machine learning

April 18th, 2023

Overview of experimental design and study workflow. The Caco-2 cells were grown in 96-well plate until 70% monolayer. Then broth was changed, in half of wells the camptothecin (a topoisomerase inhibitor) was added and cultivation was followed for 24 h. Then cells were stained with either Acridine Orange (3 μg/ml) or DioC6 (0.02 μg/ml) and propidium iodide (3 μg/ml) and series of 3 wells were analyzed with either microscopy on Carl Zeiss Observer 1.0 microscope with 40× magnification or with flow cytometry using BD FACSCanto II flow cytometer. Credit: Kazan Federal University

A paper by KFU scholars saw light in Scientific Data.

The formed microimage database was used in the framework of complex research together with specialists of the Saint Petersburg State Electrotechnical University (LETI), who developed a program complex which allows using artificial intelligence algorithms from microphotographs to quantify cell death by cell structure under the influence of advanced drugs without the need to use a staining process. Thus, it is possible to automatically exclude obviously unsuccessful samples of candidates for medicinal compounds in real time without exerting additional influence on the cells, which avoids the use of expensive dyes, but at the same time to assess the effect of the drug on cells within a given period of time.

Chair of the Department of Genetics Airat Kayumov comments on the KFU-LETI cooperation, "Since 2017, we have been working closely with them in the field of biomedical image processing. In 2018, they published a program for semi-automatic analysis of images with differential coloring. This year, an article was published on counting the area of cells in their monolayer. The idea came from working together to create a tool to estimate the percentage of dead cells by machine learning. To do this, we formed a microphoto base, and the colleagues implemented the algorithm."

The cells were grown by Associate Professor Pavel Zelenikhin, microscopy was managed by Senior Research Associate Elena Trizna (Laboratory of Natural Antimicrobial Preparations), and Research Associate of the same lab Diana Baidamshina.

The images model various outcomes of drug use which were obtained through two differing staining techniques. The neural network exposed to such dataset is able to differentiate outcomes without the use of fluorescent markers. The resulting algorithms can reduce workload and improve the effectiveness of analysis in oncological therapy and other instances of medication use.

More information:
Brightfield vs Fluorescent Staining Dataset–A Test Bed Image Set for Machine Learning based Virtual Staining

www.nature.com/articles/s41597-023-02065-7

Provided by Kazan Federal University