Indexing the Event Calculus: Towards practical human-readable Personal Health Systems

Area of Science:

• Computer Science
• Medical Informatics
• Artificial Intelligence

Background:

• Personal Health Systems (PHS) generate vast amounts of patient data, particularly for chronic non-communicable diseases like Type 1 Diabetes.
• Current PHS struggle with efficient data indexing and querying of physiological time series, hindering accurate patient monitoring by medical professionals.
• Integrating domain expertise into PHS is crucial for effective clinical decision-making, but requires adaptable knowledge representation.

Purpose of the Study:

• To develop a standardized, logic programming-based framework for expressing and utilizing domain knowledge in PHS.
• To propose and evaluate novel indexing mechanisms for enhancing the performance of Event Calculus reasoning in PHS.
• To compare the efficiency of different tree-based indexing structures (k-d trees, interval trees, red-black trees) for real-time physiological data analysis.

Main Methods:

• An Event Calculus-based reasoning framework was developed to standardize domain knowledge into monitoring rules.
• Three tree indexing structures (k-d trees, interval trees, red-black trees) were implemented and integrated into the Event Calculus Knowledge Base.
• Performance analysis involved measuring rule checking times with increasing event data volumes and varying inter-arrival times.
• The Multi-Agent paradigm was employed using the MAGPIE platform to integrate reasoning engines and sensors.

Main Results:

• The proposed Event Calculus framework effectively standardizes domain knowledge for PHS applications.
• Customized jREC demonstrated superior performance with high-frequency data (short inter-arrival times).
• Standard Event Calculus with k-d trees proved more efficient for sparse data (longer inter-arrival times).
• The Multi-Agent system successfully integrated reasoning and sensing components.

Conclusions:

• Logic programming, specifically Event Calculus, offers a robust approach for integrating expert knowledge into PHS.
• Optimized indexing strategies are essential for improving the real-time performance of PHS, with method selection dependent on data characteristics.
• The developed framework and indexing techniques enhance the capability of PHS for efficient chronic disease monitoring.