FPM2Stain Net: physics-guided super-resolution and multi-modal virtual staining for digital histopathology.

Area of Science:

• Digital Pathology
• Computational Imaging
• Biomedical Optics

Background:

• Conventional histopathology relies on chemical staining, which is time-consuming and costly.
• Digital pathology aims to enhance diagnostic capabilities through computational methods.
• Fourier ptychographic microscopy (FPM) offers label-free imaging but requires high-resolution reconstruction.

Purpose of the Study:

• To develop an integrated computational pipeline (FPM2Stain Net) for high-resolution, multi-modal digital histopathology.
• To enable physics-guided super-resolution and deep learning-based virtual staining.
• To provide a cost-effective and scalable alternative to chemical staining.

Main Methods:

• Utilized bidirectional physics-based Fourier ptychographic microscopy (BiP-FPM) with a self-supervised ResNet-U-Net for high-resolution reconstruction.
• Employed a multi-task conditional generative adversarial network (cGAN) for synthesizing virtual stains (H&E, DAPI, LAP2, panCK).
• Incorporated wavelet-based spatial-frequency fusion and perceptual supervision for enhanced synthesis accuracy.

Main Results:

• FPM2Stain Net demonstrated superior reconstruction fidelity and staining accuracy compared to conventional FPM, GAN-based, and diffusion-based methods.
• Synthesized virtual stains preserved fine structural details and improved downstream analyses like cell segmentation and biomarker quantification.
• Achieved a >10× pixel-level upsampling factor from low-magnification input.

Conclusions:

• FPM2Stain Net offers a fast, scalable, and cost-effective solution for digital pathology.
• The pipeline enables high-resolution, multi-modal imaging without chemical staining.
• This technology has potential applications in multiplex imaging and point-of-care diagnostics.