Related Experiment Video
Updated: Nov 29, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
836
Generalized Fixation Invariant Nuclei Detection Through Domain Adaptation Based Deep Learning
IEEE Journal of Biomedical and Health Informatics
|November 19, 2020
Summary
Sample fixation methods significantly impact deep learning nucleus detection accuracy in histology. Understanding this variability is crucial for developing robust automated nucleus detection algorithms for diverse histopathological images.
Area of Science:
- Digital pathology
- Computational biology
- Histopathology
Background:
- Nucleus detection is vital for histological image analysis.
- Histological image variability from sample preparation challenges automated nucleus detection.
Purpose of the Study:
- To investigate the impact of histopathological sample fixation methods on deep learning-based nucleus detection accuracy.
- To evaluate convolutional neural networks (CNNs) performance across different fixation types.
Main Methods:
- Trained deep learning models using hematoxylin and eosin (H&E) stained images from three fixation methods: PAXgene, formalin, and frozen.
- Assessed nucleus detection accuracy of various CNN architectures.
- Utilized a dataset of over 67,000 annotated nuclei from 16 patients.
Main Results:
- Sample preparation variability, specifically fixation methods, significantly affects the generalization capability of nucleus detection models.
- Different CNNs exhibited varying sensitivities to fixation-induced image variations.
Conclusions:
- Histopathological sample fixation is a critical factor influencing automated nucleus detection model performance.
- Developing robust nucleus detection algorithms requires accounting for sample preparation variability.
- Unsupervised domain adaptation can enhance model generalization to unseen domains, including different tissues and labs.
