Enhancing ontology-driven diagnostic reasoning with a symptom-dependency-aware Naïve Bayes classifier

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Knowledge Representation

Background:

• Ontologies are crucial in academia and industry, but often lack uncertainty reasoning capabilities.
• Existing medical ontologies lack vital probabilistic medical knowledge essential for decision-making.
• Electronic Medical Records (EMR) contain rich, yet underexplored, probabilistic medical knowledge.

Purpose of the Study:

• To discover and integrate medical knowledge probabilities from EMR texts into existing ontologies.
• To develop a novel classifier for accurate diagnostic probability assessment.
• To enrich medical ontologies with reliable probability information for improved applications.

Main Methods:

• Building a medical ontology by identifying entities from EMRs.
• Proposing a symptom-dependency-aware naïve Bayes classifier (SDNB) to model symptom interdependencies.
• Incorporating disease probabilities into the ontology using innovative techniques.

Main Results:

• Experiments on over 30,000 deidentified EMRs identified probabilities for 31 out of 336 gastrointestinal diseases.
• Successfully integrated 31 disease probabilities and 189 conditional probabilities into the ontology.
• Validated the method's ability to discover meaningful and accurate medical knowledge probabilities.

Conclusions:

• The proposed method effectively extracts and integrates medical knowledge probabilities from EMR data into ontologies.
• The SDNB classifier accurately calculates disease probabilities, enhancing diagnostic capabilities.
• Combining ontologies with the SDNB classifier offers significant advantages for medical applications.