Generating reference models for structurally complex data. Application to the stabilometry medical domain.

Area of Science:

• Medical informatics
• Data science
• Biostatistics

Background:

• Structurally complex data, common in medical domains, presents challenges for traditional analysis.
• Hierarchical data structures with diverse attributes, including time series, require specialized modeling.
• Generating population reference models is crucial for medical tasks like diagnosis and patient evolution analysis.

Purpose of the Study:

• To present a novel framework for generating reference models from structurally complex, hierarchically organized data.
• To develop methods for measuring similarity, detecting outliers, and modeling time series within complex datasets.
• To support critical medical applications including diagnosis, fraud detection, and patient outcome analysis.

Main Methods:

• A conceptual model for hierarchical representation of structurally complex data was developed.
• A similarity tree concept was employed for quantifying individual data similarity.
• An event-based analysis approach was integrated for time series modeling within the framework.

Main Results:

• The framework demonstrated high accuracy in outlier detection (97.8%) in stabilometry data.
• Reference model generation achieved high classification accuracy (91.7% for athletes, 96.6% with control groups).
• Validation in the complex medical field of stabilometry confirmed the framework's efficacy.

Conclusions:

• The developed framework provides a robust solution for analyzing complex, hierarchical medical data.
• The methods for outlier detection and reference model generation are validated by high performance metrics.
• This approach supports the creation of population archetypes for advanced medical data analysis.