Related Experiment Video
Updated: Sep 16, 2025
08:40
Quantitation of Protein Expression and Co-localization Using Multiplexed Immuno-histochemical Staining and Multispectral Imaging
Published on: April 8, 2016
12.9K
Nuclei segmentation and classification from histopathology images using federated learning for end-edge platform
Anjir Ahmed Chowdhury1,2, S M Hasan Mahmud3,2, Md Palash Uddin4,5
1Department of Computer Science, University of Houston, Houston, Texas, United States of America.
Plos One
|July 10, 2025
Summary
This study introduces a deep learning framework for automated nuclei segmentation and classification in histology images, improving cancer detection accuracy. The efficient, privacy-preserving model shows high performance, aiding clinical diagnosis.
Area of Science:
- Digital pathology
- Computational oncology
- Artificial intelligence in medicine
Background:
- Accurate nuclei segmentation and classification in histology images are crucial for cancer detection.
- Manual methods are time-consuming and labor-intensive, necessitating automated solutions.
- Challenges include color inconsistency, blurry boundaries, and overlapping nuclei in histopathology images.
Purpose of the Study:
- To develop and evaluate a deep learning framework for enhanced nuclei segmentation and classification in histopathology images.
- To improve the accuracy and efficiency of automated cancer diagnosis through nuclei analysis.
- To ensure data privacy and enable edge device deployment.
Main Methods:
- A two-stage deep learning framework combining SegNet for segmentation and DenseNet121 for classification.
- Hyperparameter optimization using the Hyperband method.
- Federated learning (FedAvg) for decentralized, privacy-preserving training and full integer quantization for efficient deployment.
Main Results:
- SegNet achieved 91.4% Mean Pixel Accuracy (MPA), 63% Mean Intersection over Union (MIoU), and 90.6% Frequency-Weighted IoU (FWIoU).
- DenseNet121 classifier achieved 83% accuracy and 67% Matthews Correlation Coefficient (MCC), outperforming state-of-the-art models.
- Post-quantization, models showed performance gains of 1.3% and 1.0%, respectively, demonstrating efficiency.
Conclusions:
- The proposed deep learning framework accurately and efficiently segments and classifies nuclei in histopathology images.
- The privacy-preserving and quantized model is suitable for real-world clinical deployment in cancer diagnosis.
- This approach offers a scalable and automated solution to enhance cancer detection in digital pathology.
