Image-based phenotyping of disaggregated cells using deep learning

Samuel Berryman,Jeong Hyun Lee,Simon P Duffy,Kerryn Matthews,Hongshen Ma

doi:10.1038/s42003-020-01399-x

Samuel Berryman, Jeong Hyun Lee + Show 3 more

Open Access

https://doi.org/10.1038/s42003-020-01399-x

Copy DOI

Abstract

The ability to phenotype cells is fundamentally important in biological research and medicine. Current methods rely primarily on fluorescence labeling of specific markers. However, there are many situations where this approach is unavailable or undesirable. Machine learning has been used for image cytometry but has been limited by cell agglomeration and it is currently unclear if this approach can reliably phenotype cells that are difficult to distinguish by the human eye. Here, we show disaggregated single cells can be phenotyped with a high degree of accuracy using low-resolution bright-field and non-specific fluorescence images of the nucleus, cytoplasm, and cytoskeleton. Specifically, we trained a convolutional neural network using automatically segmented images of cells from eight standard cancer cell-lines. These cells could be identified with an average F1-score of 95.3%, tested using separately acquired images. Our results demonstrate the potential to develop an “electronic eye” to phenotype cells directly from microscopy images.

Highlights

The ability to phenotype cells is fundamentally important in biological research and medicine
The watershed algorithm was used on each candidate image patch to determine if multiple objects were present in the nuclear or cytoplasm channels; these were rejected if multiple objects were detected as only single-cell images were desired
We investigated whether deep-learning could phenotype single cells directly from microscopy images that are unidentifiable to the human eye

Summary

Introduction

The ability to phenotype cells is fundamentally important in biological research and medicine. Immunofluorescence phenotyping may be undesirable because: (1) phenotyping markers may be unavailable or lack specificity, (2) the sample may be too heterogeneous, (3) number of markers required may exceed the number of available fluorescence channels that can be detected, and (4) specific labeling may affect the cell in undesirable ways, such as activation or loss of viability In many of these situations, an important question is whether individual cells could be phenotyped directly using microscopy images without specific labeling. Phenotyping cells with more subtle morphologies has been largely restricted to binary classification in order to detect specific alterations resulting from disease[10,14,15,16,17] Another approach is to use brightfield microscopy images to predict the location of immune-stains on sub-cellular structures in order to identify organelles[18,19,20,21]. A further step is required to interpret these stains to establish the cell phenotype

Methods

Results

Conclusion