A mutual neighbor-based clustering method and its medical applications

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Clustering analysis is crucial for real-world applications.
• K-means is a popular clustering technique due to its simplicity.
• K-means performance is sensitive to initial centers and struggles with manifold clusters.

Purpose of the Study:

• To propose a novel clustering algorithm capable of identifying arbitrary-shaped clusters.
• To address the limitations of traditional K-means, including sensitivity to initial centers and inability to detect manifold structures.

Main Methods:

• Utilizes mutual neighbors to estimate data object density.
• Identifies high-density representative objects.
• Introduces path distance, derived from a minimum spanning tree, to measure similarities for manifold structures.
• Improves K-means by incorporating representative objects and path-based distances for initial centers and similarity measurement.
• Assigns cluster labels to non-representative objects based on neighborhood information.

Main Results:

• The proposed clustering method effectively identifies arbitrary-shaped clusters.
• Demonstrates superior performance compared to state-of-the-art clustering methods in experiments on synthetic data.
• Successfully applied to medical scenarios, identifying distinct disease types.

Conclusions:

• The novel clustering algorithm overcomes K-means limitations in identifying complex cluster shapes.
• The method provides a robust approach for density estimation and similarity measurement in clustering.
• The algorithm shows significant potential for applications requiring the identification of arbitrary-shaped clusters, including medical data analysis.