Light scattering imaging modal expansion cytometry for label-free single-cell analysis with deep learning

Area of Science:

• Biomedical Imaging
• Computational Biology
• Cellular Analysis

Background:

• Single-cell imaging is crucial for drug development, disease diagnosis, and personalized medicine.
• Acquiring multi-modal information from label-free cells presents a significant challenge.
• Existing methods often require labels or provide limited data from single-cell imaging.

Purpose of the Study:

• To develop a novel method, modal expansion cytometry, for label-free single-cell analysis.
• To expand single-mode light scattering images into multi-modal images (bright-field and fluorescence).
• To enhance the diagnostic and analytical capabilities in single-cell research.

Main Methods:

• Utilized a deep learning architecture to convert single-mode light scattering images into multi-modal representations.
• Employed a novel network optimization method combining adversarial loss, L1 distance loss, and VGG perceptual loss.
• Validated the method using simulated data, standard spheres, and various cell types, including cancer and leukemia cells.

Main Results:

• Expanded bright-field and fluorescence images closely matched conventional microscopy results.
• Achieved high contour ratio accuracy (near 1) for both whole cells and nuclei.
• Demonstrated improved cervical cancer cell subtyping accuracy (92.85%) compared to single-mode imaging.

Conclusions:

• Modal expansion cytometry effectively generates artificial multimodal images from label-free single-cell light scattering data.
• The technique provides enhanced cell visualization and improved cell classification capabilities.
• Shows significant potential for applications in single-cell analysis, including cancer diagnosis.