To enhance the reliability of pseudo-labeling in semi-supervised medical image classification by leveraging contextual relationships in data, particularly in ambiguous cases.
Key Findings:
HSC achieved 92.24% accuracy on NCT-CRC-HE with only 200 labeled samples, outperforming PEFAT by 2.97%, demonstrating its effectiveness in limited data scenarios.
HSC reached 94.17% accuracy on ISIC2018 with 20% labeled data, showing a 2.21% improvement over existing methods, indicating its potential for broader applications.
Interpretation:
The proposed HSC framework effectively utilizes contextual information in medical images to improve the reliability of pseudo-labeling, which is crucial for accurate diagnosis in clinical practice.
Limitations:
The framework's performance may vary with different types of medical images and datasets, such as MRI versus CT scans.
Further validation is needed to assess generalizability across diverse medical imaging tasks, including different diseases and imaging modalities.
Conclusion:
HSC represents a significant advancement in semi-supervised medical image classification by integrating multi-level contextual relationships for improved learning.
