Automated labeling in medical data: A semi-supervised density-based approach for efficient diagnosis model...

Area of Science:

• Medical data analysis
• Machine learning in healthcare
• Computational biology

Background:

• Automated diagnosis and analysis models are crucial for healthcare practitioners due to the increasing volume of medical data.
• Manual data labeling for machine learning models is expensive, time-consuming, and prone to errors.
• Current methods struggle with the efficient utilization of large unlabeled medical datasets.

Purpose of the Study:

• To enhance semi-supervised learning performance by automating the data labeling process.
• To reduce the cost and complexity associated with developing automated medical diagnosis models.
• To improve the accuracy of diagnostic models using limited labeled data.

Main Methods:

• Developed a novel Semi-Supervised Density Based Clustering with a confidence region (SSDCAR) algorithm.
• Automated data labeling by identifying peak density samples and constructing clusters from unlabeled data.
• Analyzed sample distribution within clusters to identify high and low confidence regions for label propagation.

Main Results:

• The proposed SSDCAR algorithm demonstrated superior performance compared to existing methods on benchmark medical datasets.
• Achieved a significant accuracy increase of at least 2% across multiple health datasets.
• The algorithm proved to be scalable for larger datasets and memory-efficient.

Conclusions:

• SSDCAR offers a more accurate and efficient approach to semi-supervised learning in medical data analysis.
• The method effectively reduces the cost and manual effort required for data labeling.
• This technique provides a scalable and computationally efficient solution for developing automated diagnostic tools.