Deep-learning-based cross-modality translation from Stokes image to bright-field contrast

Area of Science:

• Biomedical optics
• Computational pathology
• Deep learning in microscopy

Background:

• Mueller matrix (MM) microscopy offers subwavelength microstructural insights but is not standard clinical practice.
• Pathologists rely on bright-field microscopy of stained tissues for disease diagnosis.
• Cross-modality translation using polarization imaging can improve pathological analysis.

Purpose of the Study:

• To develop a deep learning technique for translating snapshot Stokes images into bright-field microscopy contrast.
• To enable bright-field visualization from polarization data without requiring MM images.
• To create a method robust to variations in light source and samples.

Main Methods:

• Utilized CycleGAN, a deep learning model, for image translation, eliminating the need for co-registered training image pairs.
• Employed snapshot Stokes images as input for the translation model.
• Applied the method to stained pathological tissue slides of liver, breast, and lung tissues.

Main Results:

• Successfully generated bright-field equivalent images from Stokes images across different staining styles.
• Demonstrated the method's effectiveness on various tissue types (liver, breast, lung) and staining protocols (H&E, TTF-1, Ki-67).
• Evaluated output image quality using four standard assessment methods.

Conclusions:

• Stokes images, acquired faster and independent of light intensity and registration, can be effectively translated to bright-field images.
• The proposed computational technique enhances the utility of polarization imaging for pathological diagnosis.
• This approach offers a stable and efficient alternative for analyzing tissue microstructures.