Related Experiment Video
Updated: Aug 19, 2025
10:25
Deep Learning-Based Segmentation of Cryo-Electron Tomograms
Published on: November 11, 2022
9.2K
A dual decoder U-Net-based model for nuclei instance segmentation in hematoxylin and eosin-stained histological
Amirreza Mahbod1,2, Gerald Schaefer3, Georg Dorffner4
1Institute for Pathophysiology and Allergy Research, Medical University of Vienna, Vienna, Austria.
Frontiers in Medicine
|November 28, 2022
Summary
This study introduces a novel dual decoder U-Net model for precise nuclei instance segmentation and classification in histological images. The method enhances diagnostic accuracy in digital pathology by automating nuclei analysis.
Area of Science:
- Digital Pathology
- Computational Biology
- Medical Image Analysis
Background:
- Microscopic analysis of tissue sections, particularly nuclei characteristics, is vital for cancer diagnosis and treatment decisions.
- Digital pathology and automated image analysis offer faster, more objective alternatives to manual microscopic evaluation.
- Supervised convolutional neural networks (CNNs) represent the state-of-the-art for automated nuclei detection, segmentation, and classification.
Purpose of the Study:
- To develop and evaluate a CNN-based dual decoder U-Net model for accurate nuclei instance segmentation in H&E-stained histological images.
- To implement an independent U-Net model for nuclei classification, complementing the segmentation task.
- To assess the model's performance on publicly available datasets for nuclei analysis in digital pathology.
Main Methods:
- A CNN-based dual decoder U-Net architecture was designed for nuclei instance segmentation, predicting foreground and distance maps.
- A watershed algorithm and post-processing refinements were employed to generate final instance segmentation masks.
- A separate U-Net model was developed for nuclei classification based on the segmentation output.
Main Results:
- The proposed dual decoder U-Net model achieved high performance in nuclei instance segmentation across three datasets.
- Average panoptic quality scores reached 50.8% (CryoNuSeg), 51.3% (NuInsSeg), and 62.1% (MoNuSAC).
- The model secured the top rank on the MoNuSAC dataset's post-challenge leaderboard, demonstrating its effectiveness.
Conclusions:
- The developed dual decoder U-Net model provides a robust and accurate solution for nuclei instance segmentation and classification in digital pathology.
- Automated analysis of nuclei in histological images significantly aids human experts, improving efficiency and objectivity.
- This approach holds promise for advancing precision medicine through enhanced diagnostic capabilities in histopathology.
