Duplex to Monoplex IHC Image Translation with Auxiliary CycleGAN Guidance
Core Concepts
Duplex to monoplex IHC image translation is improved through the use of auxiliary CycleGAN guidance, addressing the ambiguous mapping challenge.
Abstract
1. Abstract:
Generative models like Cycle GANs are used for domain translation.
The translation between duplex and monoplex IHC images poses challenges due to non-bijective mapping.
An alternative approach using auxiliary IF images is proposed for better translation results.
2. Introduction:
CycleGANs are efficient but face challenges in translating between different IHC domains.
A novel training design is introduced to address the limitations and improve translation results.
3. Methods:
Illustration of the method with domains A, B, and C for duplex, monoplex, and IF images respectively.
Loss functions and guidance mechanisms explained for effective domain translation.
4. Results:
Dataset details and quantitative evaluation results presented.
Proposed method shows significant improvements in downstream segmentation tasks compared to baselines.
5. Discussion:
Novel stain translation algorithm resolves mapping ambiguity between different IHC images.
Use of auxiliary IF domain enables better guidance for direct translation, outperforming CycleGAN baseline.
References: Listed references supporting the research and methodology used.
Auxiliary CycleGAN-guidance for Task-Aware Domain Translation from Duplex to Monoplex IHC Images
Stats
Cycle Generative Adversarial Networks (GANs) are well established.
Quantitative and qualitative results on a downstream segmentation task show benefits of the proposed method.
The duplex IHC dataset consists of 16 whole slide images (WSI).
Monoplex IHC and IF datasets consist of 35K and 4.5K unlabeled patches respectively.
Quotes
"An alternative constrain leveraging a set of immunofluorescence (IF) images as an auxiliary unpaired image domain."
"We propose - through the introduction of a novel training design, an alternative constrain leveraging a set of immunofluorescence (IF) images as an auxiliary unpaired image domain."