MINE: An Interactive Platform for Expert-Guided Medical Information Extraction.

Area of Science:

• Health Informatics
• Natural Language Processing
• Clinical Data Management

Background:

• Clinical reports are rich in patient data but challenging to analyze due to unstructured text and privacy concerns.
• Existing methods for extracting information from clinical notes are often limited in scope or require extensive manual effort.
• Developing automated, yet accurate, information extraction tools is crucial for leveraging clinical data for research and patient care.

Purpose of the Study:

• To introduce MINE, a novel, secure, and interactive platform designed for collaborative development, evaluation, and deployment of clinical information extraction models.
• To integrate diverse extraction methodologies, including rule-based, BERT-based, and large language models (LLMs), into a cohesive framework.
• To demonstrate the effectiveness of expert-guided model development for accurate clinical data extraction.

Main Methods:

• The MINE platform was designed as a unified environment integrating multiple information extraction techniques.
• Experts collaboratively developed, trained, and validated models within the secure MINE interface.
• The platform supports rule-based, BERT-based, and LLM-based approaches, allowing for flexible model creation.

Main Results:

• Initial evaluations focused on extracting data for smoking status, diabetes, and hypertension.
• The expert-guided extraction models developed on MINE achieved high performance, with F-scores exceeding 95% for these conditions.
• These results validate the feasibility and effectiveness of the MINE platform for clinical information extraction.

Conclusions:

• MINE offers a viable solution for overcoming the challenges of unstructured clinical data and privacy constraints.
• Expert-guided development within the MINE platform enables the creation of highly accurate clinical information extraction models.
• The platform's success in early trials suggests significant potential for improving clinical data utilization in healthcare and research.