Inpainting Missing Tissue in Multiplexed Immunofluorescence Imaging.

Area of Science:

• Biomedical Imaging
• Computational Pathology
• Digital Pathology

Background:

• Multiplex immunofluorescence (MxIF) enables simultaneous staining of multiple markers on a single tissue section.
• Repeated staining and bleaching in MxIF can lead to physical depletion of scarce tissue.
• Limited tissue availability restricts downstream single-cell analysis.

Purpose of the Study:

• To investigate the feasibility of using generative adversarial networks (GANs) to synthesize missing tissue in MxIF imaging.
• To develop a multi-channel, high-resolution image synthesis approach for reconstructing depleted tissue regions.
• To quantitatively evaluate the performance of synthesized tissues in downstream applications like cell membrane segmentation.

Main Methods:

• Employed generative adversarial network (GAN) approaches for image synthesis.
• Integrated a multi-channel, high-resolution image synthesis technique.
• Utilized 11 structural molecular markers (epithelial and stromal) for synthesis.
• Quantitatively evaluated synthesis performance using cell membrane segmentation as a downstream task.

Main Results:

• The proposed GAN-based synthesis method showed comparable reproducibility to baseline methods for reconstructing missing tissue regions.
• The synthesis method achieved a 40% improvement in whole tissue synthesis, crucial for practical applications.
• Quantitative evaluation via cell membrane segmentation demonstrated the effectiveness of the synthesized tissues.

Conclusions:

• Generative adversarial networks (GANs) show promise for synthesizing missing tissues in MxIF imaging.
• Deep image synthesis can enhance the utility of MxIF data by increasing usable tissue areas.
• This approach offers a digital solution to tissue depletion challenges in multiplex immunofluorescence.