DAWN: Domain-Adaptive Weakly Supervised Nuclei Segmentation via Cross-Task Interactions

TL;DR

DAWN introduces a domain-adaptive weakly supervised nuclei segmentation framework that leverages cross-task interactions between segmentation and a point-annotated detection task to enable effective domain transfer. It combines a Consistent Feature Constraint to align feature spaces, a Combined Pseudo-label optimization to fuse segmentation and detection outputs, and Interactive Supervision Training for progressive, iterative labeling guidance. Across six datasets, DAWN outperforms existing weakly supervised methods and achieves parity with or surpasses several fully supervised methods, demonstrating robust cross-domain performance from PanNuke and MoNuSeg sources. The work highlights practical impact for scalable, annotation-efficient nuclei segmentation in computational pathology and provides a code release for reproducibility.

Abstract

Weakly supervised segmentation methods have gained significant attention due to their ability to reduce the reliance on costly pixel-level annotations during model training. However, the current weakly supervised nuclei segmentation approaches typically follow a two-stage pseudo-label generation and network training process. The performance of the nuclei segmentation heavily relies on the quality of the generated pseudo-labels, thereby limiting its effectiveness. This paper introduces a novel domain-adaptive weakly supervised nuclei segmentation framework using cross-task interaction strategies to overcome the challenge of pseudo-label generation. Specifically, we utilize weakly annotated data to train an auxiliary detection task, which assists the domain adaptation of the segmentation network. To enhance the efficiency of domain adaptation, we design a consistent feature constraint module integrating prior knowledge from the source domain. Furthermore, we develop pseudo-label optimization and interactive training methods to improve the domain transfer capability. To validate the effectiveness of our proposed method, we conduct extensive comparative and ablation experiments on six datasets. The results demonstrate the superiority of our approach over existing weakly supervised approaches. Remarkably, our method achieves comparable or even better performance than fully supervised methods. Our code will be released in https://github.com/zhangye-zoe/DAWN.

Figures (11)

• Figure 1: The weakly supervised nuclei segmentation flows based on point annotation. (a) represents a traditional two-stage segmentation paradigm. The point annotation is employed to generate the pseudo label, which is then used to train the segmentation network. (b) represents our domain-adaptive weakly supervised segmentation method. The point annotation is utilized to train a detection network, which can assist target domain transfer.
• Figure 2: The problems of generated pseudo labels. (a) represents the input image; (b) represents the generated mask using Voronoi qu2019weakly; (c) represents the generated mask using LSM zhang2022ddtnet; (d) is the ground-truth label of input.
• Figure 3: The schematics of our proposed DAWN framework. (a) represents the pre-trained segmentation network in source domain; (b) expresses the generation of extended point annotation using Gaussian distribution; (c) and (d) represent target domain transfer process.
• Figure 4: The schematic diagram of combined pseudo-label optimization. (a) is probability output of detection network; (b) is detection mask through filtering; (c) is segmentation mask; (d) is combined map of detection and segmentation outputs; (e) is point annotation; (f) is generated pseudo-label.
• Figure 5: The visualization comparison between weakly supervised methods and our proposed DAWN.