An Affordable and Versatile Two-Dimensional Cell Isolation and Tracking Platform Based on Image Machine Learning and Maskless Photolithography Single Cell Encapsulation

An Affordable and Versatile Two-Dimensional Cell Isolation and Tracking Platform Based on Image Machine
Learning and Maskless Photolithography Single Cell Encapsulation
Current commercial cell sorters typically use sheath flow to align cells into a single profile and sort cells based
on fluorescence signal or images. The single profile alignment limits the throughput and requires complex
hardware and expensive equipment for high-speed sorting. The usage of high-speed sheath flow also generates
high stress on cells, which makes it not suitable for fragile or sensitive cells such as stem cells for downstream
application. Some sticky cells such as monocytes or too many dead cells in the sample can interrupt or even
clog the flow. Such cell sorter also usually requires a significant amount of starting cell number. Considering the
yield, purity, and fluid dead volume, it is challenging to sort out cells of rare population such as subset of stem
cells or circulating tumor cells in blood sample. There are strong needs from small labs for an affordable and
versatile cell sorting platform applicable to a variety of cell types. The objectives of the proposed work are to: 1.
Develop a high-speed machine learning-based cell classification module. The module will enable real-time
detection of target cells inside a wide microfluidic channel based on brightfield or fluorescent images. 2. Develop
a stop flow lithography-based 2D cell sorting platform in combination with acoustic field cell array patterning that
will generate encoded encapsulations of target cells of different sizes. 3. Integrate the machine learning detection
and maskless lithography with the size-based filtering/sorting of the cell into an affordable cell sorter. The setup
can be mounted onto existing microscope and high-resolution camera, along with a web-lab flow controller and
a UV projector, makes a versatile and affordable cell sorter.
The proposed method can sort multiple cell types based on high content image information and machine learning.
This eliminates the dependency on specific antibody types which is the basis of fluorescence-activated cell
sorting (FACS) or magnetics-activated cell sorting (MACS). The proposed method can use simple microfluidic
devices for sorting different types of target cells in high purity with minimum requirement on starting cell number,
thus is applicable to rare subset of a large sample or rare cells. Maskless lithography based on digital micromirror
device (DMD) is used to stamp encoded ID to track individual cells which is convenient for downstream analysis.
The 2D wide platform can avoid high shear flow-induced cell damage or property change in the cell sorting
channel, thus is suitable for gentle cells such as stem cells. The wide channel can also avoid the potential cell
clogging problem in a regular cell sorter. By updating the machine learning algorithm and sharing datasets and
pre-trained models, as well as the availability of cameras and projectors of better resolution, the proposed project
leads to an affordable, expandable, powerful, and universal cell sorting platform.

Grant Number: 5R21EB033102-02
NIH Institute/Center: NIH
Principal Investigator: YEVGENY BERDICHEVSKY