Unpaired Image-to-Image Translation for Segmentation and Signal Unmixing

2025-05-27

AI Summary

This work addresses unpaired cross-domain image style transfer for biomedical images, proposing the Ui2i model to achieve high-fidelity style translation while preserving structural content integrity. Methodologically, it is the first to leverage real unpaired immunohistochemistry (IHC) and immunofluorescence (IF) images for nucleus segmentation domain adaptation and unsupervised single-channel IF signal disentanglement. Key innovations include approximate bidirectional spectral normalization, a U-Net–based generator with skip connections, and a channel-spatial joint attention mechanism, integrated within an enhanced CycleGAN framework. Experiments demonstrate that Ui2i significantly improves segmentation accuracy in IHC domain adaptation and, for the first time, enables unsupervised disentanglement of overlapping biological signals in IF images. To our knowledge, this is the first method successfully achieving interpretable, structure-preserving disentanglement of superimposed IF signals on real unpaired data—establishing a novel, biomedically grounded paradigm for explainable image translation.

Technology Category

• Computer Vision: Medical and Biological Imaging
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Humanities & Computational Social Science

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Web Mining and Content Analysis: Bridging structured and unstructured data

Abstract

This work introduces Ui2i, a novel model for unpaired image-to-image translation, trained on content-wise unpaired datasets to enable style transfer across domains while preserving content. Building on CycleGAN, Ui2i incorporates key modifications to better disentangle content and style features, and preserve content integrity. Specifically, Ui2i employs U-Net-based generators with skip connections to propagate localized shallow features deep into the generator. Ui2i removes feature-based normalization layers from all modules and replaces them with approximate bidirectional spectral normalization -- a parameter-based alternative that enhances training stability. To further support content preservation, channel and spatial attention mechanisms are integrated into the generators. Training is facilitated through image scale augmentation. Evaluation on two biomedical tasks -- domain adaptation for nuclear segmentation in immunohistochemistry (IHC) images and unmixing of biological structures superimposed in single-channel immunofluorescence (IF) images -- demonstrates Ui2i's ability to preserve content fidelity in settings that demand more accurate structural preservation than typical translation tasks. To the best of our knowledge, Ui2i is the first approach capable of separating superimposed signals in IF images using real, unpaired training data.

Problem

• Enables style transfer across domains while preserving content integrity
• Improves disentanglement of content and style features in unpaired image translation
• Separates superimposed signals in immunofluorescence images using unpaired data

Innovation

• U-Net generators with skip connections
• Bidirectional spectral normalization for stability
• Channel and spatial attention mechanisms