Deep learning enabled label-free microfluidic droplet classification for single cell functional assays.

Area of Science:

• Biotechnology
• Microfluidics
• Machine Learning

Background:

• Droplet-based microfluidics and microscopy enable single-cell analysis but generate large datasets.
• Classifying droplets by cell number is crucial but visually analyzing images is time-consuming.
• Existing machine learning models struggle with droplets containing non-cellular structures alongside cells.

Purpose of the Study:

• To develop a deep learning model for accurate and rapid classification of cells in droplet-based microfluidic assays.
• To address the challenge of classifying droplets containing both cells and non-cellular structures.
• To create a generalized model applicable to diverse microfluidic applications.

Main Methods:

• Development of a deep learning model utilizing the ResNet-50 neural network architecture.
• Application of the model to functional droplet-based microfluidic assays for image analysis.
• Training and validation of the model on microscopy images of droplets.

Main Results:

• The deep learning model achieved >90% accuracy in classifying droplets based on cell count.
• The model demonstrated high accuracy in classifying droplets with both cells and non-cellular structures, as well as cells alone.
• The model showed generalization capabilities across different cell types and microfluidic applications.

Conclusions:

• The developed deep learning model significantly accelerates the analysis of droplet-based microfluidic assays.
• This approach provides a robust solution for classifying cell-containing droplets, even in complex samples.
• The model offers a broadly applicable tool for enhancing the efficiency and accuracy of single-cell analysis in microfluidics.