Artificial Intelligence-Driven Structurization of Diagnostic Information in Free-Text Pathology Reports

Area of Science:

• Medical Informatics
• Natural Language Processing (NLP)
• Artificial Intelligence (AI)

Background:

• Pathology report free-text contains crucial diagnostic information.
• Current computer-based analytics underutilize this data.
• Existing NLP methods struggle with complex entities and relationships.

Purpose of the Study:

• To develop a novel informatics pipeline for extracting complex diagnostic entities and relationships.
• To transform extracted information into Knowledge Graphs (KGs) of relational triples (RTs).
• To enable advanced data-mining applications in diagnostic medicine.

Main Methods:

• Extended open information extraction (openIE) techniques.
• Integrated artificial intelligence (AI) based modeling.
• Structured output as Knowledge Graphs (KGs) of relational triples (RTs).

Main Results:

• High semantic similarity (Mean Weighted Overlap 0.83) with original reports.
• High precision (0.925) and recall (0.841) for extracted RTs.
• Significant inter-rater agreement (93.6%) and reliability (81.8%).

Conclusions:

• The pipeline demonstrates high accuracy, minimality, and adequate knowledge representation.
• It effectively extracts complex diagnostic entities and relationships.
• The pipeline is suitable for various downstream data-mining applications to aid diagnostic medicine.