Abstract Objective: To develop a comprehensive approach for identifying live cell nuclei in images without fluorescent labels. Since cell biology involves counting cells, assessing cell growth dynamics, and confluence, it is expedient to automate the collection of these data. Machine learning algorithms are used for automation and must be trained on images of specific cell cultures. Training algorithms is a labor-intensive process and requires lengthy manual annotation. Additionally, available machine learning-based analysis methods have low accuracy in identifying living cells without fluorescent staining. Materials and methods: The methodology involved the use of convolutional neural networks based on an algorithm for segmenting cell nuclei in fluorescent and histological images using StarDist. To create annotated phase-contrast images of cell cultures, samples were stained with the nuclear fluorescent dye DAPI, followed by the rejection of poor-quality images using classification in the CellProfiler Analyst program. The StarDist-based model was trained on 1 130 images of automatically annotated nuclei in phase-contrast images of human respiratory tract epithelial cell cultures, obtained with a 10× lens, 1 600 × 1 200 pixels in size, and 16-bit grayscale depth. Results: The resulting model showed good accuracy (F1 = 0.765) in segmenting nuclei on the validation dataset. The model was used to determine the population doubling time of the epithelial cell culture. Conclusion: The developed approach made it possible to create annotations and train a machine learning model to obtain data without the use of fluorescent labels (“label-free”) on live cell cultures.
Balyasin et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: