Label-free cell classification in holographic flow cytometry through an unbiased learning strategy

Area of Science:

• Biomedical Engineering
• Computational Biology
• Optical Imaging

Background:

• Label-free imaging flow cytometry is crucial for single-cell analysis in diagnostics and life sciences.
• Digital holographic microscopy offers multi-refocusing and quantitative phase imaging for rich sample information.
• Deep learning enhances cell classification but faces generalization challenges due to data biases.

Purpose of the Study:

• To develop a robust deep learning framework for label-free cell classification using digital holographic microscopy.
• To overcome data biases in holographic imaging settings for improved model generalization.
• To accurately classify drug-resistant endometrial cancer cells.

Main Methods:

• A Mask R-CNN model for cell detection.
• A convolutional auto-encoder for label-free feature extraction, mitigating experimental biases.
• A feedforward neural network for single-cell classification based on extracted features.

Main Results:

• The proposed hybrid deep learning model successfully classifies cells using label-free holographic imaging.
• The approach demonstrates effective feature extraction from unlabelled data, overcoming experimental setting biases.
• Accurate identification of drug-resistant endometrial cancer cells was achieved in a challenging classification task.

Conclusions:

• The integrated deep learning framework enhances the reliability of label-free cell classification in digital holographic microscopy.
• This method paves the way for developing generalized, bias-resilient diagnostic tools for point-of-care applications.
• The approach holds significant potential for advancing single-cell analysis in clinical diagnostics and healthcare.