Hybrid Personal Medical Digital Assistant Agents

Autonomous intelligent systems are beginning to impact clinical practice as personal medical assistant agents, by leveraging experts’ knowledge when needed and exploiting the vast amount of patient data available to clinicians. However, these approaches are seldom integrated. In this paper, we propose an integrated hybrid agent architecture that combines symbolic reasoning with sub-symbolic, data-driven models. Using the PIMA dataset, we demonstrate that this hybrid approach enhances the performance of both approaches when used alone. Specifically, we show that integrating a logical agent, which uses predefined expert knowledge plans, with rules obtained by symbolic knowledge extraction from machine learning models trained on historical data, improves system reliability and clinical decision-making, while reducing misclassified instances.