Leukocyte differentiation in bronchoalveolar lavage fluids using higher harmonic generation microscopy and deep...

Area of Science:

• Medical imaging
• Computational pathology
• Immunology

Background:

• Interstitial lung diseases (ILDs) diagnosis relies on analyzing bronchoalveolar lavage fluid (BALF) and biopsies.
• Standard leukocyte differentiation in BALF is labor-intensive and time-consuming.
• Third harmonic generation (THG) and multiphoton excited autofluorescence (MPEF) microscopy show promise for leukocyte identification in blood.

Purpose of the Study:

• Extend leukocyte differentiation to BALF samples using THG/MPEF microscopy.
• Develop a deep learning algorithm for automated leukocyte identification and quantification in BALF.

Main Methods:

• Isolated leukocytes from blood and BALF samples.
• Imaged cells using label-free THG/MPEF microscopy.
• Trained a deep learning model on 2D images to estimate leukocyte ratios.

Main Results:

• Identified distinct leukocyte populations in BALF using label-free microscopy.
• Deep learning network achieved >90% accuracy in estimating leukocyte percentages on BALF samples.
• Demonstrated distinctive cytological characteristics for neutrophils, eosinophils, lymphocytes, and macrophages.

Conclusions:

• Label-free THG/MPEF microscopy and deep learning enable rapid leukocyte differentiation and quantification.
• This approach can accelerate ILD diagnosis, decrease costs, and reduce inter-observer variability.
• Automated analysis of BALF offers significant potential for clinical diagnostics.