A Light-Weight Text Summarization System for Fast Access to Medical Evidence

Area of Science:

• Medical Informatics
• Natural Language Processing
• Scientific Literature Analysis

Background:

• The rapid growth of medical research, exemplified by the COVID-19 pandemic, makes staying current with evidence challenging for professionals.
• Existing automatic summarization systems often require extensive domain knowledge and resources, limiting their speed and deployability.

Purpose of the Study:

• To develop a fast, simple, and easily deployable extractive summarization approach for medical research.
• To enable medical experts and researchers to rapidly access the latest research evidence.

Main Methods:

• The proposed approach utilizes similarity measurements from pre-trained medical domain-specific word embeddings.
• It incorporates simple features and avoids computationally expensive pre-processing or resource-heavy knowledge bases.
• The system performs query-focused extractive summarization at runtime.

Main Results:

• Automatic evaluation using ROUGE on an evidence-based medicine dataset showed statistically comparable performance to state-of-the-art methods.
• Manual evaluation on recent COVID-19 articles indicated performance close to human agreement levels for extractive summarization.

Conclusions:

• The developed summarization approach offers a fast and simple solution for accessing up-to-date medical research.
• Its ease of deployment and efficiency make it suitable for medical experts and researchers, especially in rapidly evolving fields.