Patient clustering using dynamic partitioning on correlated and uncertain biomedical data

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Physiological Data Analysis

Background:

• Health professionals analyze multiple physiological data to identify treatment patterns for clinical abnormalities.
• Biomedical data often reveals common patterns associated with specific clinical illnesses.
• Discovering these patterns in vital sign data can aid in identifying patients with similar conditions.

Purpose of the Study:

• To develop an automated method for discovering patterns in multi-dimensional vital sign data using unsupervised learning.
• To utilize discovered patterns for identifying and grouping patients with similar clinical conditions.
• To introduce a novel patient clustering approach based on aggregated instance-wise uncertainty (AIU).

Main Methods:

• A patient clustering method is developed to group patients by discovering dynamic patterns in multi-dimensional vital sign data.
• A dynamic partitioning algorithm is proposed, incorporating aggregated instance-wise uncertainty (AIU) computed from physiological time-series data.
• The method employs unsupervised learning for pattern discovery and patient grouping.

Main Results:

• The developed model was evaluated using principal component analysis and silhouette values.
• Quantitative evaluation demonstrated the model's ability to cluster patients with different clinical situations.
• Experiments on real-world data achieved accuracies of 82.85% and 91.17% on two distinct datasets, outperforming state-of-the-art methods.

Conclusions:

• The proposed approach effectively discovers distinct patterns in vital sign data with predictive significance.
• The patient clustering method demonstrates high accuracy in grouping patients based on physiological data patterns.
• This technique offers a promising tool for identifying similar clinical conditions through unsupervised learning.