Core Concepts
Proposing an inter- and intra-uncertainty regularization method for improved histopathology image segmentation.
Abstract
The content introduces a novel feature aggregation model for semi-supervised histopathology image segmentation. It addresses the limitations of existing methods by incorporating inter- and intra-uncertainty regularization strategies. The proposed PG-FANet model outperforms state-of-the-art methods on MoNuSeg and CRAG datasets, showcasing competitive performance with limited labeled data.
Directory:
Introduction
Importance of accurate instance segmentation in histology images.
Challenges in manual segmentation due to complex morphological features.
Existing Methods in SSL for Biomedical Image Segmentation
Categories: consistency regularization, pseudo label generation, adversarial learning, contrastive learning.
Limitations of previous SSL methods in histopathology image segmentation.
Proposed Method: PG-FANet Architecture
Two-stage network with MGFE and MMFA modules for feature aggregation.
Uncertainty Estimation and Consistency Regularization
Inter- and intra-uncertainty modeling to address prediction discrepancies.
Experimental Results and Comparison with State-of-the-Art Models on MoNuSeg and CRAG datasets.
Performance Metrics: F1-score, Dice coefficient, IoU, AJI, 95HD for nuclei segmentation; Diceobj, Hausobj, 95HDobj for gland segmentation.
Stats
"Various semi-supervised learning approaches have been developed to work with limited ground truth annotations."
"To address these challenges existing in fully- or semi-supervised histopathology image segmentation..."
"Our PG-FANet outperforms other state-of-the-art methods..."
"Our proposed semi-supervised learning framework yields competitive performance with a limited amount of labeled data."