CyclePose -- Leveraging Cycle-Consistency for Annotation-Free Nuclei Segmentation in Fluorescence Microscopy

2025-03-14

AI Summary

To address the scarcity of annotated data for nucleus instance segmentation in fluorescence microscopy images, this paper proposes the first end-to-end unsupervised framework that jointly performs synthetic image generation and segmentation. Innovatively, a differentiable segmentation head is embedded within a CycleGAN architecture, enabling self-training without paired image-mask data or manual annotations—guided by pixel-level cycle-consistency constraints, adversarial loss, and structure-preserving regularization. Unlike conventional two-stage “generate-then-segment” paradigms, the integrated design significantly reduces computational overhead. Evaluated on two public benchmark datasets, the method outperforms existing weakly supervised and unsupervised approaches, achieving segmentation accuracy comparable to fully supervised U-Net under zero-label conditions. This work establishes a novel, efficient, and practical unsupervised segmentation paradigm for biomedical image analysis.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Unsupervised & Self-Supervised Learning
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Security and Privacy: Data transparency and provenance

Abstract

In recent years, numerous neural network architectures specifically designed for the instance segmentation of nuclei in microscopic images have been released. These models embed nuclei-specific priors to outperform generic architectures like U-Nets; however, they require large annotated datasets, which are often not available. Generative models (GANs, diffusion models) have been used to compensate for this by synthesizing training data. These two-stage approaches are computationally expensive, as first a generative model and then a segmentation model has to be trained. We propose CyclePose, a hybrid framework integrating synthetic data generation and segmentation training. CyclePose builds on a CycleGAN architecture, which allows unpaired translation between microscopy images and segmentation masks. We embed a segmentation model into CycleGAN and leverage a cycle consistency loss for self-supervision. Without annotated data, CyclePose outperforms other weakly or unsupervised methods on two public datasets. Code is available at https://github.com/jonasutz/CyclePose

Problem

• Addresses nuclei segmentation without annotated datasets.
• Reduces computational cost by integrating data generation and segmentation.
• Outperforms weakly or unsupervised methods on public datasets.

Innovation

• CyclePose integrates synthetic data generation and segmentation training.
• Uses CycleGAN for unpaired image-to-mask translation.
• Leverages cycle consistency loss for self-supervised learning.