Weak-to-Strong Generalization Enables Fully Automated De Novo Training of Multi-head Mask-RCNN Model for Segmenting Densely Overlapping Cell Nuclei in Multiplex Whole-slice Brain Images

2025-12-12

AI Summary

This work addresses the unsupervised segmentation of densely overlapping nuclei in multiplexed cyclic immunofluorescence (mCIF) whole-slide images—a challenging task requiring no manual annotations. We propose a fully automated, annotation-free training framework for multi-head Mask R-CNN. Methodologically, we introduce a novel “weak-to-strong” generalization paradigm integrating efficient channel attention, iterative pseudo-label correction with coverage expansion, unsupervised domain adaptation, and a self-supervised segmentation quality assessment mechanism—enabling zero-shot cross-device and cross-protocol transfer. Evaluated against five state-of-the-art methods, our approach achieves significant performance gains. To foster reproducibility and clinical translation, we publicly release the source code, sample data, and high-resolution segmentation results.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Security and Privacy: Data transparency and provenance

Abstract

We present a weak to strong generalization methodology for fully automated training of a multi-head extension of the Mask-RCNN method with efficient channel attention for reliable segmentation of overlapping cell nuclei in multiplex cyclic immunofluorescent (IF) whole-slide images (WSI), and present evidence for pseudo-label correction and coverage expansion, the key phenomena underlying weak to strong generalization. This method can learn to segment de novo a new class of images from a new instrument and/or a new imaging protocol without the need for human annotations. We also present metrics for automated self-diagnosis of segmentation quality in production environments, where human visual proofreading of massive WSI images is unaffordable. Our method was benchmarked against five current widely used methods and showed a significant improvement. The code, sample WSI images, and high-resolution segmentation results are provided in open form for community adoption and adaptation.

Problem

• Automated training for overlapping nuclei segmentation
• Learning new image classes without human annotations
• Self-diagnosis metrics for segmentation quality in production

Innovation

• Weak-to-strong generalization for automated training
• Multi-head Mask-RCNN with efficient channel attention
• Automated self-diagnosis metrics for segmentation quality