Related Experiment Videos
Deep feature optimization using fusion of multiple self-supervised learning approaches and filter-based feature
Nepolian Vailankanni1, Bharanidharan Nagarajan1
1School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, Vellore, India.
Plos One
|April 27, 2026
Summary
This study enhances lung cancer detection using a novel combination of self-supervised learning and graph convolutional networks. The new method significantly improves diagnostic accuracy from histopathological images.
Area of Science:
- Oncology
- Computer Science
- Medical Imaging
Background:
- Lung cancer is a leading cause of mortality globally, necessitating early and accurate detection.
- Current diagnostic methods for lung cancer from imaging are often subjective, highlighting the need for advanced computational techniques.
- Deep learning offers potential for computer-aided diagnosis, but accuracy remains a critical challenge.
Purpose of the Study:
- To improve the accuracy of lung cancer diagnostics from histopathological images.
- To develop an optimized feature fusion strategy by combining multiple self-supervised learning techniques.
- To leverage Vision Graph Convolutional Networks for enhanced classification.
Main Methods:
- Utilized a custom Convolutional Neural Network for initial feature extraction from histopathological images.
- Applied three self-supervised learning methods (Deep Cluster, Bootstrap Your Own Latent, Simple Framework for Contrastive Learning of Visual Representations) to refine features.
- Integrated Minimum Redundancy Maximum Relevance for filter-based feature selection and employed Vision Graph Convolutional Networks for classification.
Main Results:
- Achieved a balanced accuracy of 97% on the LungHist700 dataset, a significant improvement over the baseline Vision Graph Convolutional Network (86%).
- Validated the proposed approach on LC25000 and TCGA UT datasets, demonstrating consistent enhanced lung cancer prediction performance.
- The optimized feature fusion strategy effectively combined complementary strengths from different self-supervised learning approaches.
Conclusions:
- The proposed method, integrating multiple self-supervised learning techniques, feature selection, and Vision Graph Convolutional Networks, substantially enhances lung cancer diagnostic accuracy.
- This approach offers a promising direction for developing more reliable computer-aided diagnostic tools for lung cancer.
- The findings underscore the potential of advanced deep learning strategies in improving early detection and patient outcomes for lung cancer.