Deep learning-based 3D leukocyte differentiation using label-free higher harmonic generation microscopy.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Cell Biology

Background:

• Leukocyte differentiation is crucial for diagnosing diseases and understanding pathophysiology.
• Current methods like manual cytospins are slow, labor-intensive, and operator-dependent.
• There is a need for automated, high-throughput methods for leukocyte analysis.

Purpose of the Study:

• To develop and validate a deep learning framework for rapid, automated 3D leukocyte differentiation.
• To utilize label-free higher harmonic generation microscopy (HHGM) for imaging.
• To compare the performance of deep learning models (ResNet 3D-50, ViT 3D) against gold-standard methods.

Main Methods:

• Performed 3D leukocyte imaging using label-free HHGM on bronchoalveolar lavage fluid (BALF) and blood samples.
• Trained and tested ResNet 3D-50 and Vision Transformer (ViT) 3D models for leukocyte differentiation.
• Compared deep learning predictions with manual cytospin analyses using Bland-Altman analysis.

Main Results:

• Achieved >86% accuracy for BALF and >96% accuracy for blood samples with the deep learning framework.
• Demonstrated close agreement with gold-standard cytological analysis, with mean differences <5% across leukocyte subpopulations.
• The framework provided rapid, automated 3D imaging and differentiation.

Conclusions:

• Integrated label-free HHGM with deep learning for fast, accurate, high-throughput leukocyte differentiation.
• The proposed technology significantly improves efficiency and reproducibility in analyzing BALF and blood samples.
• This approach has the potential to revolutionize clinical workflows and advance precision medicine.