Leveraging Rule-Based NLP to Translate Textual Reports as Structured Inputs Automatically Processed by a Clinical...

Area of Science:

• Computational linguistics and medical informatics.
• Application of Natural Language Processing (NLP) in oncology.

Background:

• Clinical Decision Support Systems (CDSSs) require structured patient data for effective breast cancer management.
• Extracting this data from unstructured clinical reports is challenging, and current machine learning methods often act as black boxes.

Purpose of the Study:

• To develop and evaluate a rule-based NLP method for automating the extraction of breast cancer patient data.
• To translate unstructured patient summaries into structured profiles for guideline-based CDSS input within the DESIREE project.

Main Methods:

• Implementation of a rule-based NLP pipeline involving Named Entity Recognition (NER), relation extraction, and structured data extraction.
• Systematic organization of patient data into structured profiles.
• Comparison of CDSS recommendations generated from NLP-extracted data versus manually created patient profiles (gold standard).

Main Results:

• The rule-based NLP method demonstrated strong alignment with treatment recommendations, showing only a 2% difference compared to the gold standard.
• The NER pipeline achieved high performance: an average F1-score of 0.9 for main entities (patient, side, tumor).
• Relation extraction achieved an F1-score of 0.87, and contextual information extraction achieved 0.75.

Conclusions:

• Rule-based NLP offers a transparent and effective approach for structuring breast cancer patient data from clinical reports.
• The developed method shows promising results for integration into CDSSs, potentially enhancing breast cancer management.
• The high performance metrics indicate the robustness of the rule-based NLP approach for clinical text analysis.