A Semisupervised Learning Scheme with Self-Paced Learning for Classifying Breast Cancer Histopathological Images
Sarpong Kwadwo Asare1, Fei You1, Obed Tettey Nartey2
1School of Electronic Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China.
Computational Intelligence and Neuroscience
|December 30, 2020
Summary
This study introduces a novel semisupervised learning method for breast cancer image classification. It effectively uses unlabeled data and handles class imbalance, improving classifier efficiency.
Area of Science:
- Medical Imaging
- Machine Learning
- Computational Pathology
Background:
- Medical imaging tasks face challenges due to limited labeled data and expert labeling requirements.
- Data imbalance further complicates the training of machine learning algorithms in medical diagnostics.
Purpose of the Study:
- To develop a robust semisupervised learning method for breast cancer histopathological image classification.
- To address data scarcity and imbalance issues in medical image analysis.
Main Methods:
- Integration of self-training and self-paced learning for pseudolabel generation and selection.
- A novel algorithm generates and selects high-confidence pseudolabels, prioritizing less-represented classes.
- A class balancing framework normalizes confidence scores to manage data imbalance.
Main Results:
- The proposed method effectively utilizes unlabeled data to augment limited labeled datasets.
- Improved classification performance was demonstrated on the BreakHis dataset.
- The approach successfully handles data imbalance, preventing the neglect of underrepresented classes.
Conclusions:
- The developed semisupervised learning method offers a promising solution for breast cancer image classification with limited labeled and imbalanced data.
- The novel pseudolabeling and class balancing strategies enhance classifier efficiency and robustness.
