Label-Free White Blood Cell Classification Using Refractive Index Tomography and Deep Learning.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Hematology

Background:

• Conventional blood cell analysis methods are time-consuming, costly, and require specialized expertise.
• Existing label-free imaging techniques for blood cells can be complex and slow.

Purpose of the Study:

• To develop a rapid and accurate blood cell identification method using deep learning and label-free refractive index (RI) tomography.
• To enable quantitative analysis of morphological and biochemical properties of bone marrow white blood cells (WBCs).

Main Methods:

• Acquisition of RI tomograms of bone marrow WBCs using a Mach-Zehnder interferometer-based tomographic microscope.
• Classification of WBCs using a 3D convolutional neural network.
• Quantitative parameter extraction directly from tomograms.

Main Results:

• Achieved >99% accuracy for binary classification (myeloids vs. lymphoids).
• Achieved >96% accuracy for four-type classification (monocyte, myelocyte, B lymphocyte, T lymphocyte).
• Visualized feature learning capabilities using unsupervised dimension reduction.

Conclusions:

• The proposed method offers a cost-effective and rapid approach for blood cell classification.
• This framework can be integrated into existing workflows for improved hematologic malignancy diagnosis.
• Deep learning with RI tomography provides a powerful tool for hematology research.