An End-to-End Platform for Digital Pathology Using Hyperspectral Autofluorescence Microscopy and Deep Learning-Based...

Area of Science:

• Digital Pathology
• Biomedical Imaging
• Artificial Intelligence in Medicine

Background:

• Conventional histopathology is costly, time-consuming, and consumes tissue samples.
• Existing staining methods introduce variability and limit further analyses.
• There is a need for efficient, cost-effective, and reproducible pathology workflows.

Purpose of the Study:

• To develop an end-to-end automated pathology workflow using hyperspectral microscopy and deep learning.
• To generate virtual histologic stains from unstained tissue autofluorescence.
• To enable automated disease scoring and facilitate spatial biology research.

Main Methods:

• Custom hyperspectral microscope developed for nondestructive imaging of unstained tissue autofluorescence.
• Deep learning models trained to generate virtual stains from autofluorescence data.
• Nonalcoholic steatohepatitis (NASH) scoring comparison study using real and virtual stains.
• Development of deep learning models for automated NASH Clinical Research Network score prediction.

Main Results:

• Virtual stains accurately reproduced histologic features seen in real-stained images.
• Moderate-to-good concordance observed between pathologists' scoring of real and virtual stains.
• Automated pathology platform demonstrated comparable performance to expert pathologists for NASH scoring.

Conclusions:

• Hyperspectral microscopy combined with deep learning offers a viable virtual staining strategy.
• This approach enhances cost-efficiency, reliability, and tissue sample conservation in pathology.
• The platform shows potential for improving disease diagnosis and advancing spatial biology research.