Accurate Classification of Human CD4+ T, CD8+ T, and CD19+ B Cells Isolated from Splenocytes by Cross-Polarized...

Area of Science:

• Biophysics
• Cell Biology
• Machine Learning

Background:

• Diffraction imaging offers rapid cellular phenotyping based on molecular responses to light.
• Distinguishing diverse human leukocyte types using diffraction imaging requires further investigation.

Purpose of the Study:

• To investigate the capability of polarization diffraction imaging to classify human lymphocyte subtypes.
• To develop and validate a deep neural network for analyzing polarization diffraction imaging data.

Main Methods:

• Cross-polarized diffraction image (p-DI) pairs from live human lymphocytes were acquired.
• A dual-channel deep neural network (DNN), named DINet-PS, was developed for feature extraction and filtering in the angular frequency domain.
• The network incorporated adaptive spectral filters to mitigate noise in p-DI pairs.

Main Results:

• The DINet-PS model achieved a classification accuracy of 96.6 ± 0.40% on hold-out test data.
• Accurate classification was demonstrated for three major lymphocyte subtypes: CD4+ T, CD8+ T, and CD19+ B cells.
• The DNN effectively extracted cell-specific features from p-DI data.

Conclusions:

• Deep neural networks can successfully extract relevant cellular features from polarization diffraction imaging data.
• Polarization diffraction imaging flow cytometry shows significant potential for label-free classification of lymphocyte and leukocyte subtypes.