Related Experiment Video
Updated: Dec 11, 2025
12:56
An Automated Method to Perform The In Vitro Micronucleus Assay using Multispectral Imaging Flow Cytometry
Published on: May 13, 2019
23.0K
Pollen analysis using multispectral imaging flow cytometry and deep learning
Susanne Dunker1,2, Elena Motivans2,3,4, Demetra Rakosy1,2
1Helmholtz-Centre for Environmental Research - UFZ, Permoserstraße 15, Leipzig, 04318, Germany.
The New Phytologist
|August 18, 2020
Summary
This study introduces a novel method combining multispectral imaging flow cytometry and deep learning for rapid and accurate pollen identification and quantification. This automated approach achieves 96% accuracy, surpassing manual microscopy for diverse research applications.
Area of Science:
- Botany
- Ecology
- Evolutionary Biology
- Bioinformatics
Background:
- Pollen analysis is vital for ecological, evolutionary, and applied research but manual microscopy is time-consuming and challenging.
- Automated methods are sought to improve efficiency and accuracy in pollen identification and quantification.
- Manual microscopy remains the standard despite ongoing research into alternative techniques.
Purpose of the Study:
- To develop and validate a novel method for automated pollen analysis using multispectral imaging flow cytometry and deep learning.
- To achieve high-accuracy, rapid identification and quantification of pollen grains from various plant species.
- To enable detailed extraction of pollen morphological traits for further analysis.
Main Methods:
- Utilized multispectral imaging flow cytometry for high-throughput pollen data acquisition.
- Developed and trained a convolutional neural network classifier on a dataset of 426,876 pollen images from 35 plant species.
- Integrated deep learning for automated species identification, pollen quantification, and morphological trait analysis.
Main Results:
- Achieved a species-averaged accuracy of 96% in pollen identification using the deep learning classifier.
- Demonstrated successful differentiation of even morphologically similar pollen species.
- Enabled accurate quantification of pollen grains and detailed extraction of traits like size, symmetry, and structure.
- Phylogenetic analyses indicated potential phylogenetic conservatism in certain pollen traits.
Conclusions:
- The developed method offers a powerful, rapid, and accurate tool for pollen analysis, suitable for diverse research needs.
- This approach significantly enhances pollen identification, quantification, and trait extraction compared to traditional methods.
- A comprehensive pollen reference database can further optimize the application of this technology in various scientific fields.
