Transformer-based active learning for multi-class text annotation and classification.

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Data-driven healthcare relies on labeled data, but manual annotation of unstructured clinical notes is challenging.
• Lack of explicit labels in medical data hinders effective decision-making and analysis.

Purpose of the Study:

• To develop a novel deep active learning framework for efficient multiclass text classification of clinical notes.
• To automate the annotation process using the SOAP (subjective, objective, assessment, plan) framework.

Main Methods:

• Leveraged transformer-based deep learning models for automatic annotation of clinical notes.
• Implemented a deep active learning framework to facilitate the annotation process.

Main Results:

• Achieved superior classification accuracy on a diverse set of over 426 clinical notes.
• Demonstrated an F1 score improvement of 4.8% over existing methods.
• Validated the practical utility for healthcare professionals and clinical documentation.

Conclusions:

• The synergy between active learning and deep learning advances automatic text annotation in clinical informatics.
• Future work will explore multimodal data and large language models for enhanced clinical text analysis.