A model for indexing medical documents combining statistical and symbolic knowledge.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Knowledge Representation

Background:

• Accurate indexing of medical documents is crucial for information retrieval and clinical decision-making.
• Existing methods may lack contextual understanding and precision in diverse documentary settings.

Purpose of the Study:

• To develop and evaluate an information processing method for precise medical document indexing.
• To leverage both symbolic and statistical knowledge for contextual term relevance.

Main Methods:

• A model integrating general knowledge from the Unified Medical Language System (UMLS) with domain-specific statistical knowledge was designed.
• Extracted terms were ranked to identify the most significant ones within each document.
• The model was validated on 17,079 French standardized discharge summaries (SDSs).

Main Results:

• The proposed method successfully ranked the primary ICD-10 term within the top two positions in nearly 90% of cases.
• This demonstrates high accuracy in identifying the most critical medical concepts.

Conclusions:

• Utilizing multiple terminologies enhances the precision of medical document indexing.
• The integration of semantic relationships significantly improves the performance of indexing models.